{"id": "train_20000", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "governance", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5bbeead378deeb68bc7282336fff0d0b011907fe"}
{"id": "train_20001", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "documentation", "governance"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5993ad71f417e23d6dd64453be4236defd0de693"}
{"id": "train_20002", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "evaluation_metrics", "auditability", "documentation"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2059964c637caf2d68cb6c1051efcfe3cd2a4078"}
{"id": "train_20003", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "tooling", "reproducibility", "tests_are_truth"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "293282c3a7954d5cf09a605a15cb8011daf493b7"}
{"id": "train_20004", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "evaluation_metrics", "tooling"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "965780648d89ab462b2664f9209e51d9bf9dc194"}
{"id": "train_20005", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "documentation", "evaluation_metrics", "governance"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d8a666b03843fbe3e5596115b2d36e5231cf86f3"}
{"id": "train_20006", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "evaluation_metrics", "reproducibility", "tests_are_truth"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "68f630f8f6cb042281133e33f9a2774fcc9e5c12"}
{"id": "train_20007", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "ci_integration", "tooling", "documentation"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "aea77666479a011b49902461644f6f1b0356df2f"}
{"id": "train_20008", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "repo_scale_reasoning", "governance", "tooling"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "079fdc8fe3c4f403ee1ed1a1284fbe27d45cd64e"}
{"id": "train_20009", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "evaluation_metrics", "reproducibility", "governance"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0a25dd7c1ba82b1446d96d854c827dab1e7f4a04"}
{"id": "train_20010", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "tooling", "governance", "security_gates"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "28de5dab2c841a63dab75ed17e723d29fb11dbbb"}
{"id": "train_20011", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "documentation", "evaluation_metrics", "tooling"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "747e4b588811f5fcd6a9f09a3a20e5a7412b0534"}
{"id": "train_20012", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "tests_are_truth", "security_gates"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1c88172c0836998a3106a750fcf0c4737208fa8d"}
{"id": "train_20013", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "ci_integration", "governance", "security_gates"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5c86c638e5854cce751370e0c77b6b5c89b15746"}
{"id": "train_20014", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "reproducibility", "documentation", "tooling"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "30efd5b79bd76b2957ab26639b2147e36badd8ca"}
{"id": "train_20015", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "ci_integration", "cost_latency_tradeoffs", "security_gates"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5b9f7253288f843a14f66e9cda8fe154db898b43"}
{"id": "train_20016", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "tests_are_truth", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0b3d0f2199a5fe469478a1e177e6e8ccfc74b4b6"}
{"id": "train_20017", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "security_gates", "governance", "reproducibility"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "695cc85b8c2fb29fcbc3f01e76fb96c516050ace"}
{"id": "train_20018", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "auditability", "evaluation_metrics", "governance"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f309d769211b31ce082fbe9b7bc12e3c7f34da7b"}
{"id": "train_20019", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "tooling", "evaluation_metrics"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "499a9c207db59c5d920a6954b543967e3d6531f3"}
{"id": "train_20020", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "tooling", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "256c1773d7875675bd15bc01bf0891b3c81f5cdb"}
{"id": "train_20021", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "auditability", "governance", "reproducibility"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ee63f681409f67ea3ac605de89d644a14e865432"}
{"id": "train_20022", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "tests_are_truth", "documentation", "governance"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a86da61187b5c0caab8701754a357ca827a782cc"}
{"id": "train_20023", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "evaluation_metrics", "security_gates", "tooling"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e8474beaf6f86e7175ef8fb9fb769bf0afdca8d4"}
{"id": "train_20024", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "tooling", "repo_scale_reasoning", "reproducibility"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "56373181aced1235c0ef94b3449072131791a646"}
{"id": "train_20025", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "security_gates", "tooling", "tests_are_truth"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b68f7f72a45cfa93c55eaba80c563ab58a29e43b"}
{"id": "train_20026", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "tests_are_truth", "governance", "documentation"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "79b50f5cfbc7ce8f47ce5fc9de160b5b9ae69bff"}
{"id": "train_20027", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "governance", "auditability", "tests_are_truth"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9650cc91be34a06fab046edca6fa5fa339041471"}
{"id": "train_20028", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "auditability", "tooling", "governance"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "05bd7681089f817e7de4a3379b5e0323b2d60110"}
{"id": "train_20029", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "documentation", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9bdff2309885852a13e64664af60ffaf19bfd17d"}
{"id": "train_20030", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "evaluation_metrics", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b3965cb7bdd0d25d404097f36142d3075b1eb33e"}
{"id": "train_20031", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "documentation", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "df6bdae11dfdeb5cf260ac710af6c7ec6f82a73f"}
{"id": "train_20032", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "governance", "ci_integration", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6b31e82d50d1144f062b552c81cbee98505881d4"}
{"id": "train_20033", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "tooling", "tests_are_truth", "auditability"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e545ad87131d453293ca1888f31b924b01ae7ce3"}
{"id": "train_20034", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "auditability", "documentation", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8e26c884f4b4436c7bc38d2422fbe2c2ebda9641"}
{"id": "train_20035", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "governance", "tooling", "security_gates"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "873f8469b20596af12a8289e2c231bf9c3ea5a5d"}
{"id": "train_20036", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["documentation", "governance", "ci_integration", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b3093956d9215923887ba740f2d40b50acf9fd7e"}
{"id": "train_20037", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "auditability", "ci_integration"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "56b404699850c3af148c0cea2c90818d9f404220"}
{"id": "train_20038", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "tooling", "evaluation_metrics", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fb53e937d8ac08d2ca312bd6af26492082048cae"}
{"id": "train_20039", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "tests_are_truth", "ci_integration", "governance"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a21cee340fd2f874de6f9eb4a4f35cb97c4a7b3b"}
{"id": "train_20040", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "governance", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3be63cd75795b58027bcda8c49c6f36779b9cbda"}
{"id": "train_20041", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "tooling", "reproducibility", "security_gates"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "67d8bd0c23b0daa6e7d796947edd9f9afd0c71c4"}
{"id": "train_20042", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "security_gates", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2a06ccda13f628819d2ec3608918e864dd109f85"}
{"id": "train_20043", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "documentation", "tooling", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a5d59b331cde9caa59d7c47315090ff8aeec6c6e"}
{"id": "train_20044", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "ci_integration", "tooling", "documentation"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "825ed42826a2d98700b74b1da6df182d6251d8ed"}
{"id": "train_20045", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "governance", "tests_are_truth"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e25a4d4a30cd113e7317af894aeee69257a51a86"}
{"id": "train_20046", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "security_gates", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8b8ee63b81962015e749cb1b8e1ec11adfe987ec"}
{"id": "train_20047", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "documentation", "governance", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1ec06510984c4ffea71fc6af10bf62ff341c0367"}
{"id": "train_20048", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "security_gates", "governance", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "97cd3224985c5c2ff2746f0e9c19ffa17b634536"}
{"id": "train_20049", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "security_gates", "repo_scale_reasoning", "reproducibility"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7b52bdbc6d423f93294a73e4a26d72ebdbc69a12"}
{"id": "train_20050", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "auditability", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "be3ef86333b7e7a9baa2325e637aff451f5a0233"}
{"id": "train_20051", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "governance", "documentation"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2ab5167126159f012fcbf45c441fb6ac1b24ffe8"}
{"id": "train_20052", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "tooling", "reproducibility"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fa0ceecd41438a8e79f494f5e08c3089238de99b"}
{"id": "train_20053", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "tooling", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ea851455f6e0c024fb3ab09b6882d22517403c20"}
{"id": "train_20054", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "tooling", "repo_scale_reasoning", "governance"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0c9c3537859bfd550a2767b69597cdededff946d"}
{"id": "train_20055", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "documentation", "security_gates"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fd446a81c46c72d65cae15c330e7466654face24"}
{"id": "train_20056", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "auditability", "security_gates", "reproducibility"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "127854640f009b147b84edbf373a38a71b5b2476"}
{"id": "train_20057", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "security_gates", "tooling"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "517cd278b78b0e8dc12bd08826ba16625f20c64c"}
{"id": "train_20058", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "governance", "evaluation_metrics", "reproducibility"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e10e5b11cd9d3c5532ebb9b3bdb0ddaf7404a54c"}
{"id": "train_20059", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "evaluation_metrics", "tooling", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "511247b312f40a356c82e7434eae62ca0bfa67b4"}
{"id": "train_20060", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "auditability", "repo_scale_reasoning"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c522f5a77cc750786632dcfd3f6699df2f48fc31"}
{"id": "train_20061", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "governance", "tooling", "ci_integration"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "97b75a62827c0053c4622a115c378df62aa51fde"}
{"id": "train_20062", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "evaluation_metrics", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3c4022c4b3fa89de13da049d7983757c7e8c30d4"}
{"id": "train_20063", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "repo_scale_reasoning", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "72753a954306959d054cf9885bc04924fd6b7f0b"}
{"id": "train_20064", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["auditability", "security_gates", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "73101921169c5ce7b7673407e4e62e000d5db79c"}
{"id": "train_20065", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "security_gates", "evaluation_metrics", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bbeeac0535c2db4b482d1c6332fa0c4b6424798d"}
{"id": "train_20066", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "security_gates", "ci_integration", "auditability"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "eabcc7df6f02c43abfa0081b1d21a6f1779275db"}
{"id": "train_20067", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "governance", "tooling"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bada5367d321c4e6de66a66dac06afddd7441ac5"}
{"id": "train_20068", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "documentation", "auditability", "ci_integration"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3edabc71f088094e4fbcda37e1c475c159c5d9e0"}
{"id": "train_20069", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "auditability", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "112ff6ad8476e8e954ab8ffa67f24bf7c4a4b14b"}
{"id": "train_20070", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "reproducibility", "documentation", "security_gates"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c1263ce093db0a9b4898c4f9dc2998aa23eedf48"}
{"id": "train_20071", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "reproducibility", "evaluation_metrics", "security_gates"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "540eba8f742735870cb183c23e34215535507d1a"}
{"id": "train_20072", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "evaluation_metrics", "tooling", "documentation"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fa72155f431139b68e2fa4d0408d21884ce85117"}
{"id": "train_20073", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "auditability", "cost_latency_tradeoffs", "security_gates"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1e881a405a0217a7857e191e4894622ce1ee6073"}
{"id": "train_20074", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "security_gates", "governance", "ci_integration"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a5e1ddd4f579f827d243fce64a6f17064229b85f"}
{"id": "train_20075", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "cost_latency_tradeoffs", "repo_scale_reasoning", "tooling"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "89fccf2f4875be8a723ecb444d48da7d103d68e3"}
{"id": "train_20076", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "governance", "security_gates", "tooling"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0f08ae9b4bf3a5eede7d292ee19c7ffd8478521b"}
{"id": "train_20077", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "tooling", "governance"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e38b8586cf0ab67988fb60799f98e8a8b74881c9"}
{"id": "train_20078", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "governance", "tooling", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d36d17606321f4b601195ffe150df8001d382da0"}
{"id": "train_20079", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "tooling", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "214c29199f262534dab640cdaf4ad2ec41752826"}
{"id": "train_20080", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "evaluation_metrics", "repo_scale_reasoning", "reproducibility"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "25d2711a2b03bc3df104b794166e3cd0b42b4b25"}
{"id": "train_20081", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "security_gates", "governance", "documentation"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "61a736c871929fa2644bd9bb79f40fb726aa8347"}
{"id": "train_20082", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "auditability", "security_gates"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2f75b9ef52f386c951da4d2369c5c4f4c2540f21"}
{"id": "train_20083", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "tests_are_truth", "security_gates"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "78a94b10d5f0b14615d0bfb9693fe0909cc7283a"}
{"id": "train_20084", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "governance", "auditability", "tooling"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1af4f662343890b3d53a2e7748efad3b8470a4ad"}
{"id": "train_20085", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "documentation", "tests_are_truth", "security_gates"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "08393376d231e9ba9951a76ebce0b18b761e7350"}
{"id": "train_20086", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["documentation", "auditability", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0c61c4a035eea44b74b4d6977a225af689a78410"}
{"id": "train_20087", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "reproducibility", "documentation", "governance"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2391777efb120d95eb0f58cb875fa84c4603b8d1"}
{"id": "train_20088", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "auditability", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3b631f1e79c51c1b23cf15b678c2224bc25da9f3"}
{"id": "train_20089", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "governance", "security_gates", "reproducibility"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "eb200d83f3a04e4d1986c4ee974a88aeb53e552f"}
{"id": "train_20090", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "documentation", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "acddbf70835006030515a04f64c0e37b9dc371f3"}
{"id": "train_20091", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "security_gates", "reproducibility"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "72e29080bccdbe3eb02e80318bbf9a56ad5c45c4"}
{"id": "train_20092", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "auditability", "evaluation_metrics", "tooling"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9884a5c6cec7417ec93d9cdd7c56ddc608829fad"}
{"id": "train_20093", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "cost_latency_tradeoffs", "reproducibility", "tooling"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fdba311d7343e5d7701e9de961a3297908ea119d"}
{"id": "train_20094", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "tests_are_truth", "documentation"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6ed5f9cd2332b7d292527f8ad7a65aa6a0ae9321"}
{"id": "train_20095", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["governance", "repo_scale_reasoning", "evaluation_metrics", "tooling"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a3eba3f35258d948fbf5de04ceae1a0ba6fb2f8b"}
{"id": "train_20096", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "reproducibility", "tooling", "ci_integration"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "808ff9396fb9f5d293e898bfc40aca343107b3dd"}
{"id": "train_20097", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "governance", "tooling", "tests_are_truth"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d0d1500829c49eabf8236be6ec005cfadee979a3"}
{"id": "train_20098", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "governance", "auditability", "security_gates"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e0ee6400e225d6e44c660300ec2e242230418281"}
{"id": "train_20099", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["governance", "ci_integration", "repo_scale_reasoning", "security_gates"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7cbda16d4faa66659e0b6161f88a6eb9cf7895eb"}
{"id": "train_20100", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "evaluation_metrics", "security_gates", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "af7ce911cd079db4d68cefe4a9f5207366dd6b4f"}
{"id": "train_20101", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "documentation", "security_gates", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a350f78821fb775d4994038d511af086f26dabc6"}
{"id": "train_20102", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "ci_integration", "auditability"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c21f11a95d030370bae8f0a834f2b56544bb3e10"}
{"id": "train_20103", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "governance", "documentation", "security_gates"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1503c668802e80c3dc3354694ae756292a663d2a"}
{"id": "train_20104", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "documentation", "repo_scale_reasoning", "reproducibility"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "132369fe9450ec4cc8cfacb4c130ee74e5c8768e"}
{"id": "train_20105", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "documentation", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fc906cd3339fc319dcde11794ff6c01882540440"}
{"id": "train_20106", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "reproducibility", "repo_scale_reasoning", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0ba2ef7d41ab9cf2b0f8ac2246f036b713d4d036"}
{"id": "train_20107", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "documentation", "evaluation_metrics"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "25c54a5b8bedb2848e61b234f4a83e9a0c4bb08f"}
{"id": "train_20108", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "documentation", "governance", "auditability"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7d1888160012f08d137d146e62ea1fc89ff12abd"}
{"id": "train_20109", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "auditability", "tooling", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9b12bf8015d31e13ae0a8b7d343cac6c6ff3ba51"}
{"id": "train_20110", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "evaluation_metrics", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "54ad60b1c888915a58c44d1f747c0ebb8b137e42"}
{"id": "train_20111", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "tooling", "documentation"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "89a8a459f54890494d7b47a227015fb11fd122cb"}
{"id": "train_20112", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "evaluation_metrics", "tooling", "auditability"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fa2927fb317c67ffa7f0ae93ba17d654751202c0"}
{"id": "train_20113", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "documentation", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f26e609befe79a90844067bc484f8ad13d23fc96"}
{"id": "train_20114", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "security_gates", "governance", "tooling"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cfae3d1500ea4cef9a8c4dd8776df58f712c44d1"}
{"id": "train_20115", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["governance", "tests_are_truth", "evaluation_metrics", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9e136dd07dee6e39f0b4556fb979edf1890c378b"}
{"id": "train_20116", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "documentation", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4f24181192847160766880f428b4337324c3efb7"}
{"id": "train_20117", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "evaluation_metrics", "documentation", "governance"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "aa7a6da1a8cb88a70e6bd1b6b229e1465958ee51"}
{"id": "train_20118", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "ci_integration", "auditability", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ebf311c977b25933a766b622688caa19b96b25c2"}
{"id": "train_20119", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "documentation", "governance"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0902f7a4a71655c868da4cc518a6a0affd55e9f9"}
{"id": "train_20120", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "ci_integration", "governance", "tooling"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a249d1ba334df64e161effb3f9e6afeebd1bca76"}
{"id": "train_20121", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "evaluation_metrics", "documentation", "security_gates"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b8cfe0d7e2afecbcec62a6a8e18ab0740e029195"}
{"id": "train_20122", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "tooling", "tests_are_truth"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d975177242a87daba267d3425f2e48d40c1a6d8e"}
{"id": "train_20123", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "repo_scale_reasoning", "documentation", "evaluation_metrics"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4b90a026cf8ba1906c75458c431006edfe231c17"}
{"id": "train_20124", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "evaluation_metrics", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "90b6c8c2c501611aff7e627829ad7a52ab2e2974"}
{"id": "train_20125", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "repo_scale_reasoning", "tooling", "security_gates"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2ba841c79c50b81843770e86753c54a7bdba18d3"}
{"id": "train_20126", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "documentation", "evaluation_metrics", "governance"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fe5d39819828f03342663ab457f22e446597aed2"}
{"id": "train_20127", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["governance", "tooling", "documentation", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "321b79ed32c900b15ad5049d6deb4d41da1a9946"}
{"id": "train_20128", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "governance", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d66db1a609395cf1dac7441afeb1c8a4cae2344a"}
{"id": "train_20129", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "auditability", "governance", "security_gates"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9d8bf05a4de3c2f97216221a9f8dc64b34c8f18f"}
{"id": "train_20130", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["governance", "documentation", "reproducibility", "tooling"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "78603d76027931318ba5ba5931653827fcb35cde"}
{"id": "train_20131", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "tests_are_truth", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "54f90d9ff562a8071946763fde64d042fc0fac87"}
{"id": "train_20132", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "auditability", "tooling", "documentation"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "aaad428bd4f9ac772a607243114c7377a2eac8ab"}
{"id": "train_20133", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "documentation", "security_gates", "evaluation_metrics"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e5259f091d33df25b851cde6ae236d31bb44480e"}
{"id": "train_20134", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["governance", "reproducibility", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "78f8654d4df938d47e2d6f806b4e99a9fafdee22"}
{"id": "train_20135", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "documentation", "security_gates", "tests_are_truth"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c0797a96fbe2f5093d7f94f3e32d953ae41b5198"}
{"id": "train_20136", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "ci_integration", "tooling", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a77817cbffcceec5bcd46f0213b00a1bf8813f8a"}
{"id": "train_20137", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "governance", "ci_integration", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dfdc02ca850dad0c269e6c96386fef74c3c02404"}
{"id": "train_20138", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "reproducibility", "evaluation_metrics", "ci_integration"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bd5eb9a3752e6814e823eb37c602ddfc36180bb7"}
{"id": "train_20139", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "reproducibility", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "44e35fd43e70fe485de38ea906295df3a69ecce8"}
{"id": "train_20140", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "governance", "tooling", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bfd2ecbbbca042eea311e87b4cae207e60ebfe80"}
{"id": "train_20141", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "cost_latency_tradeoffs", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c217de961c2f09687a2759b705db1ed5b4fd15cf"}
{"id": "train_20142", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "tests_are_truth", "ci_integration", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "adb00a00bdee7ffa40d2292a99312d3de1a08904"}
{"id": "train_20143", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "tooling", "tests_are_truth", "security_gates"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "70e10bc5e020eef0a4d733495ed27d5c80f36d2d"}
{"id": "train_20144", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "tests_are_truth", "cost_latency_tradeoffs", "governance"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5c54f853b4ae41142178e3d179f85ebbdd4a1241"}
{"id": "train_20145", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "tests_are_truth", "auditability", "reproducibility"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3f74bce55dc67cee69ea1246d53da2601dc04c16"}
{"id": "train_20146", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "tooling", "tests_are_truth", "ci_integration"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b4ee26dcdba58ad8cbafbfb293ad1f6ac34f10a0"}
{"id": "train_20147", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "documentation", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "816a98b6ae4461e0afc1a5933aa6bdd3cfd2f484"}
{"id": "train_20148", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "cost_latency_tradeoffs", "evaluation_metrics", "security_gates"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "81667b547f60eaf54634a4671061f651e407be5d"}
{"id": "train_20149", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "evaluation_metrics", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c13fd196aea4174e5be845cdd0b543d0c1ebf73b"}
{"id": "train_20150", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "ci_integration", "reproducibility", "tooling"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "38a8070ff1ab065240c2411c7923b523bd2d7f2a"}
{"id": "train_20151", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "repo_scale_reasoning", "governance"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cb9b65dcea241bdf4f23ae03cb91d53d6002868e"}
{"id": "train_20152", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "repo_scale_reasoning", "cost_latency_tradeoffs", "security_gates"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5360532ca720beb24781bd2b6885f6361849033e"}
{"id": "train_20153", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "evaluation_metrics", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3627e49bd3f0695b169ea29024ded5f1bc31eafa"}
{"id": "train_20154", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "governance", "auditability", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "307a90a0adac60620c5403b15e3ef1a1d72bec86"}
{"id": "train_20155", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "governance", "tooling", "auditability"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "086bb3fff93504c68f9cc140c7416adeb01562de"}
{"id": "train_20156", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "tooling", "governance"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "010e22b917253bbfb3f05bec67301feff768dc3e"}
{"id": "train_20157", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "security_gates", "auditability", "documentation"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "74eff549d967714bc3bc35a4161c651c7a5203ff"}
{"id": "train_20158", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "auditability", "evaluation_metrics", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4c49abc24076dce99d757a3b6b7f4d8b1fcd766c"}
{"id": "train_20159", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "auditability", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "feab74f6d5acf8a01640e0714ddb649843e3824a"}
{"id": "train_20160", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "security_gates", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e98df5e8a449b4ff1e6f5bf409370fdf7d72e811"}
{"id": "train_20161", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "security_gates", "governance", "auditability"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5124d1be4c86d756e928e1ed45753160a3d1d92e"}
{"id": "train_20162", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "reproducibility", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1a0960e7e240d93684fcb891349bb29fefba49fb"}
{"id": "train_20163", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "security_gates", "repo_scale_reasoning"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "50e6eb3c3590025da29310bd8ef6f37e86bc9788"}
{"id": "train_20164", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "ci_integration", "repo_scale_reasoning", "documentation"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5564c652b11d7c40f2d2fb0d1acea8ec16d6a9b9"}
{"id": "train_20165", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "auditability", "ci_integration", "tooling"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "23be13d0276ad9622d48abb3ea3657c9ee9b8414"}
{"id": "train_20166", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "governance", "tests_are_truth", "auditability"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "adca71d176f6a4ff3fccf5a59d03e638c551c150"}
{"id": "train_20167", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "repo_scale_reasoning", "auditability", "tooling"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9882e30b93e491e36da037a008e398225708e936"}
{"id": "train_20168", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "evaluation_metrics", "auditability"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1a0960e7e240d93684fcb891349bb29fefba49fb"}
{"id": "train_20169", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "security_gates", "tests_are_truth", "ci_integration"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "085ffc9a0e530bafde87211abf12c92a517f0546"}
{"id": "train_20170", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "ci_integration", "tooling", "reproducibility"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b3722c5051a599ebe17d3f17551d262649acc321"}
{"id": "train_20171", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "evaluation_metrics", "auditability", "tests_are_truth"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fd7a02e141931f340b6918f88765ce07e99382a0"}
{"id": "train_20172", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "tests_are_truth", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0c2290ca9218a8d3f6adde5c5ac2dafb053b9f9f"}
{"id": "train_20173", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "evaluation_metrics", "cost_latency_tradeoffs", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5b8ee55ce12265561c96727782f079ec0250bf39"}
{"id": "train_20174", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "security_gates", "repo_scale_reasoning", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3c4656c710ceac33b9c5e0f2d3404c05e787cc3c"}
{"id": "train_20175", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "tests_are_truth", "auditability", "security_gates"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fc1922fc2f958ab14617d704bed1ec7109953df2"}
{"id": "train_20176", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "tests_are_truth", "ci_integration", "security_gates"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "44d9260005e8f4dd64a5e3eec28a4cc20d85f8d1"}
{"id": "train_20177", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "security_gates", "ci_integration"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e7474d8af74b72bf09b13272240af4445af5fb33"}
{"id": "train_20178", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "ci_integration", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "619c15a21fb87ea58d6b8ba8b9990d3109d498cc"}
{"id": "train_20179", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["governance", "reproducibility", "evaluation_metrics", "tooling"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "08ac0be7af5180b962954aa4685c592179a3722f"}
{"id": "train_20180", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "reproducibility", "documentation", "auditability"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8f64aacd3a2b4d97a6d58c2fa7178b842f7ee922"}
{"id": "train_20181", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "auditability", "governance", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7f788154e02769b299a22677fdabe49d26c3029a"}
{"id": "train_20182", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "reproducibility", "security_gates", "ci_integration"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cca13bcf6835e2cbf3d86a32388453c454f7dd16"}
{"id": "train_20183", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "evaluation_metrics", "auditability", "tests_are_truth"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "409cd031f3536a6f68bcfedda2045eef4c5be501"}
{"id": "train_20184", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "governance", "ci_integration", "auditability"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "627044221d50c440a26b515feee20952d938a70d"}
{"id": "train_20185", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "security_gates", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d79cbdbd39a3ef83ec44d36ccc208b5b1a703cf5"}
{"id": "train_20186", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "auditability", "governance"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "94db57e0ab68d5727dac33b2e6c6735e6038963e"}
{"id": "train_20187", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "reproducibility", "security_gates", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a4792aa412449ffe2c8016aaeb3e19be638430e3"}
{"id": "train_20188", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "governance", "reproducibility", "tooling"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "13c654f63a68559362d7f3d4fd8bc1c429ecf9a6"}
{"id": "train_20189", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "evaluation_metrics", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c559a2d943904d0308d36cbe109ab198c9eb2056"}
{"id": "train_20190", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "evaluation_metrics", "tests_are_truth"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "849852e9627652d7365bd90ce4ac1d8cd17c485b"}
{"id": "train_20191", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "cost_latency_tradeoffs", "reproducibility", "governance"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9b12bf8015d31e13ae0a8b7d343cac6c6ff3ba51"}
{"id": "train_20192", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "documentation", "tests_are_truth", "governance"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a2ff354cdeba09e9b505260df04c09f418cc321f"}
{"id": "train_20193", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "governance", "tooling", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b674a37da16fb7f860c6a083094785e985a302a9"}
{"id": "train_20194", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "security_gates", "evaluation_metrics", "reproducibility"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "78ffae18fef77dd21cb91bf188b87610d671ebea"}
{"id": "train_20195", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "tests_are_truth", "governance", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f19c5041d1de88692c48cd32ca50e74a2a9ef3dd"}
{"id": "train_20196", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "tooling", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "104b015305e5ad45fff0c01a87c6688e4ffc4f32"}
{"id": "train_20197", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "tests_are_truth", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "618c7fb26b8bf5b487fa836241050c89587531ff"}
{"id": "train_20198", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "tooling", "documentation", "ci_integration"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d97535fec9790bf3f048eca427f58f202bf1a205"}
{"id": "train_20199", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "auditability", "cost_latency_tradeoffs", "governance"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "691f38b3d62251b2e53225b4578ba6132103199f"}
{"id": "train_20200", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "reproducibility", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0571532701c083f0e60a8e3d5d3797dcc7a354bd"}
{"id": "train_20201", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "tooling", "ci_integration"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b608634be04a17337913f4f80f5aa505b9d0432a"}
{"id": "train_20202", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "tooling", "governance", "auditability"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "87d3e062601106c0289416430f670ba35bac1212"}
{"id": "train_20203", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "repo_scale_reasoning", "documentation", "reproducibility"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5d571774cbf6c9dd50318a69ce74e2cfb3eda4c8"}
{"id": "train_20204", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "security_gates", "governance", "ci_integration"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3a8c6c0f6adf28ac248922d79d0e298656f156df"}
{"id": "train_20205", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "tests_are_truth", "repo_scale_reasoning", "governance"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "aa7f4e6cef1e5bdcffb87e7f42760b06d17cf5fa"}
{"id": "train_20206", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "evaluation_metrics", "documentation", "governance"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "749ed6708588f0537cdb48d9878d9e6d8f837b22"}
{"id": "train_20207", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "documentation", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1399d2be5e46137f37d2c819662d7775b0b17759"}
{"id": "train_20208", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "tests_are_truth", "documentation", "reproducibility"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5802a0f98c087a0c20648ae2ed809b286c4ee5f1"}
{"id": "train_20209", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "auditability", "governance", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4ec7012b56c6ae9d158b5d6338fb2f2e6a6769f1"}
{"id": "train_20210", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "reproducibility", "documentation", "security_gates"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "41508e594d4a29037cb9af4b56dae65712e7bb5c"}
{"id": "train_20211", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "repo_scale_reasoning", "ci_integration", "security_gates"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "34c5122b6ea3ebf4d89d152ec30bb3b0ba1b3602"}
{"id": "train_20212", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "security_gates", "cost_latency_tradeoffs", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b6c8036cc389404d23adfa7dbd8edda1b81f6884"}
{"id": "train_20213", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "tooling", "ci_integration", "reproducibility"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6f4d4aa18e44f85a31930d2f3ad5cd75e0dcd604"}
{"id": "train_20214", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["governance", "ci_integration", "documentation", "tooling"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "46570a4b7ae39521521f933b35dd7d7e65428ad1"}
{"id": "train_20215", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "ci_integration", "security_gates"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "41eed858eee6a9f8726302da4d65dc56803b3b63"}
{"id": "train_20216", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "tooling", "cost_latency_tradeoffs", "security_gates"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "da9c5fce130d8c35ef4ecc62c92189ed7cafe3f4"}
{"id": "train_20217", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "governance", "documentation", "tooling"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "85105b0ed13ce03c811aa18c38a41708c3c0450a"}
{"id": "train_20218", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "security_gates", "tooling", "tests_are_truth"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0fd0c78d9ec0d4546c1c2de6568df2f6ac66452c"}
{"id": "train_20219", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "repo_scale_reasoning", "ci_integration", "evaluation_metrics"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "53d5c25a770e210f2aba897e9b15ffd78e9a9544"}
{"id": "train_20220", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["governance", "ci_integration", "tests_are_truth", "documentation"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6e3017df39fba547d97bd3c6668f954455f62d17"}
{"id": "train_20221", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "tests_are_truth", "documentation", "repo_scale_reasoning"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bde8b487ae3a24009f6a997fcecbc52d3240ec16"}
{"id": "train_20222", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "documentation", "auditability"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8573db12577bc47d449d78a0044464068155ca5b"}
{"id": "train_20223", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "documentation", "security_gates"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7701138d5a33d5a498fa9b90c0bc0e8fd2978a66"}
{"id": "train_20224", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "repo_scale_reasoning", "reproducibility", "evaluation_metrics"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a09caa2f37b800df3c1fc740c001bac9fbbae498"}
{"id": "train_20225", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "tooling", "evaluation_metrics", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fe5dfd6978f507ec70deb07a0df401ffdb73a9dc"}
{"id": "train_20226", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "tooling", "repo_scale_reasoning", "reproducibility"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ede7fc5bb114ffb3a510e76ae03dae122812d496"}
{"id": "train_20227", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "documentation", "tests_are_truth", "auditability"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3f1950202571672404226cb4b67328328074e271"}
{"id": "train_20228", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "tooling", "evaluation_metrics", "ci_integration"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "00b9a6999679bdcd3c0a811b2eb4a04a203069bf"}
{"id": "train_20229", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "repo_scale_reasoning", "ci_integration", "security_gates"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "85ff536d3f5b2bbcfce3796b57596b5c50684016"}
{"id": "train_20230", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "auditability", "ci_integration", "tooling"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b7df3b409f5e8e72ed85f8b4434195d64cc602c2"}
{"id": "train_20231", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "evaluation_metrics", "reproducibility", "documentation"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9ad07bc2fc525048b29b6e420ffb79f2f72d706f"}
{"id": "train_20232", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "governance", "tooling", "security_gates"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dba62d81b4cd81d8977fc81423b091aa68c928d3"}
{"id": "train_20233", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "evaluation_metrics", "tests_are_truth", "reproducibility"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "330485acdfbce80307c363790e38ddfba30aa4e0"}
{"id": "train_20234", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "documentation", "tooling"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "81591686af2d3ca22e0b384b56b27568058401d7"}
{"id": "train_20235", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "security_gates", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5a0f12828f5bd646d47b9a400b12f3bfbb73a98f"}
{"id": "train_20236", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "ci_integration", "documentation", "auditability"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "af497c93dc082ec822c8a4aa3341a2ca6d3fce10"}
{"id": "train_20237", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "evaluation_metrics", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7591b0796f3c4eb18841ba171b3250101c20cede"}
{"id": "train_20238", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "evaluation_metrics", "tooling", "repo_scale_reasoning"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "26d903020975cf29a13d1cf4a85e90da1ed159c9"}
{"id": "train_20239", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "governance", "ci_integration", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4c75496836e82413bbb0edae9888a6f99c499ba3"}
{"id": "train_20240", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "security_gates", "repo_scale_reasoning", "tooling"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "11056e32456dbf44d1514d9a22e9cab18ac2ee49"}
{"id": "train_20241", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "ci_integration", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "41bb834c2ae04164c073261a1e68fd9b406b62c8"}
{"id": "train_20242", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "tooling", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "259f31a334945e8207be0440dc91b8d56356654c"}
{"id": "train_20243", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "governance", "documentation", "security_gates"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f268a2fd4dc135af481794d30b9a00e0a5fc40bf"}
{"id": "train_20244", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "security_gates", "reproducibility", "tooling"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "17e269b3a33a448236f9b9e0c2a79081f8104ae9"}
{"id": "train_20245", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "reproducibility", "tests_are_truth", "ci_integration"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8a2b49d7d04c4f2b0a7842e1d4cbefb167414090"}
{"id": "train_20246", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "auditability", "ci_integration", "documentation"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "35ca6a4eb258d972894e5e0bbbc01760db157d42"}
{"id": "train_20247", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "security_gates", "governance", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "099c97f6176f0a0b2105f9d5575c535a765ef9b1"}
{"id": "train_20248", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "reproducibility", "documentation"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "459fa43b6486e1e149c796025265d889b8945fbf"}
{"id": "train_20249", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "cost_latency_tradeoffs", "ci_integration", "tests_are_truth"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "802431164c1859b38eb8b1c9ae1723545a2c638e"}
{"id": "train_20250", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "evaluation_metrics", "tooling"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9e9a6ca9d19c7b9f57ffcf8ce794740b0ffa1080"}
{"id": "train_20251", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "cost_latency_tradeoffs", "tooling", "ci_integration"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d8670833baf0802203f1650142d9d98e3fc4045a"}
{"id": "train_20252", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "evaluation_metrics", "documentation", "governance"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1137d3adce98b56bee53c9205d61849703415b1e"}
{"id": "train_20253", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "auditability", "repo_scale_reasoning", "security_gates"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d41742ac6bc3fee5ab93cd19f10ca0d2cf94c3ae"}
{"id": "train_20254", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "auditability", "governance", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e2aa0c93d623f387c7126a54a14057fa7e268b11"}
{"id": "train_20255", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "auditability", "evaluation_metrics", "reproducibility"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2323b12110dee14f3e0370e774e9f2de3013ea60"}
{"id": "train_20256", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "ci_integration", "auditability"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "10e6bc6d048e41a011fa69c2b8d167c614e62b46"}
{"id": "train_20257", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "security_gates", "tests_are_truth", "auditability"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b47efe4e499a2cc3f76e4deaf49cf0118350f4f6"}
{"id": "train_20258", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "ci_integration", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ee85eabcee6e4074a97bfb4395155cdbb0c57c65"}
{"id": "train_20259", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "tests_are_truth", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8b59d1bc448e936582d93e2a782a68cdbc4e3068"}
{"id": "train_20260", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "auditability", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3902d7de1c87f7b27556b4bdb5c09e70ecc2e908"}
{"id": "train_20261", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "cost_latency_tradeoffs", "reproducibility", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3e44c8cdf9dbf638d9de5131683f9a92d4fa88d7"}
{"id": "train_20262", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "governance", "tests_are_truth", "security_gates"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "566f3c300bfa13159ba617e595c7eec4b2d8a1fc"}
{"id": "train_20263", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "documentation", "governance", "ci_integration"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9b1de3a50b7f3c7ab2d848d17b17bc06cc84d77d"}
{"id": "train_20264", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "reproducibility", "tests_are_truth", "tooling"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ce0b8f38d460f94ca5246a6f24e41b0b1c5af70d"}
{"id": "train_20265", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["governance", "security_gates", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "00b9a6999679bdcd3c0a811b2eb4a04a203069bf"}
{"id": "train_20266", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "cost_latency_tradeoffs", "governance", "tests_are_truth"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0690c7c6aba656b7c6316df3ac973a5ad3fb1ec3"}
{"id": "train_20267", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b7aa6e4c72b46fde7802a5913a1340fba147be46"}
{"id": "train_20268", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "cost_latency_tradeoffs", "governance", "documentation"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "603fe480aa500b36094a63da7b7f0097de0a4f31"}
{"id": "train_20269", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "tooling", "reproducibility"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c407232f486ee1d91866f59f3ad7d7f54eb21dd2"}
{"id": "train_20270", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "evaluation_metrics", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2ed4f64058de4833e897a20725576b0f58ef5bb3"}
{"id": "train_20271", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "evaluation_metrics", "auditability", "ci_integration"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f00da9594f3e5f1360450074e4123c8b970b774e"}
{"id": "train_20272", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "evaluation_metrics", "tooling", "auditability"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9eb4fd6514873538b5f1ca8ea4005096a20c1018"}
{"id": "train_20273", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "ci_integration", "cost_latency_tradeoffs", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1b63a964afb1cd5939bca4b2258af7bebd774881"}
{"id": "train_20274", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "tooling", "governance", "auditability"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1d2b8f0b43b7ce62ca4874c331492866b8637ed2"}
{"id": "train_20275", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "auditability", "ci_integration", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "288f29b1bd60e76339565134f7bd1c98f03ed2b2"}
{"id": "train_20276", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "reproducibility", "ci_integration", "tooling"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e2e608dac3ce1b5289b795662b529301ca3b9f42"}
{"id": "train_20277", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "governance", "tests_are_truth"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cf745f55acbaa29f2753731e4d03e7a4d13a9b53"}
{"id": "train_20278", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["governance", "tooling", "ci_integration", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3467ad3ed3bfaedf01e1f07fd00d47a07b73b580"}
{"id": "train_20279", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "tooling", "governance", "ci_integration"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "498a9d816c69025ffcf443fe9e044b44a0d31c0e"}
{"id": "train_20280", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["auditability", "repo_scale_reasoning", "documentation", "ci_integration"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "eda990850506befedd03543b52d514f557db89af"}
{"id": "train_20281", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "auditability", "tooling", "ci_integration"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5935edfd6355b4a0e1f318b19efb2a20a2b7de3c"}
{"id": "train_20282", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "security_gates", "auditability", "documentation"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "53014c05562c37fb24910a9532b3fa92e72a263a"}
{"id": "train_20283", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "evaluation_metrics", "tooling"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "916122beb4c6745ce2bcccfc4e7a518c7a54ce79"}
{"id": "train_20284", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "documentation", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a0c82154a6f2b784b896dc614f11f348f921d13d"}
{"id": "train_20285", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "auditability", "tooling", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "37e4e32d5e9ffb16143f34c28e778a3bdaab6957"}
{"id": "train_20286", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "documentation", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fe5d39819828f03342663ab457f22e446597aed2"}
{"id": "train_20287", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "security_gates", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0ef8b16abf2c38c0544b82ab54c2d06c6ba667c6"}
{"id": "train_20288", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "security_gates", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "498453613423d9a197c377119b88731ef1f32b91"}
{"id": "train_20289", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "tests_are_truth", "security_gates", "governance"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3d74931321744b25fecff8b48cb56f4857b3620a"}
{"id": "train_20290", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "cost_latency_tradeoffs", "tests_are_truth", "tooling"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5f9a1436f57856a61f06cc52bf88d4f2599e0afa"}
{"id": "train_20291", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "auditability", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2fe5a0d05aa7ee58cbadc84e45545526977c2187"}
{"id": "train_20292", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "security_gates", "governance", "tests_are_truth"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "99b1e450d6a38cf00afa7f4d9d9cb1cd472b2544"}
{"id": "train_20293", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "repo_scale_reasoning", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f684ee0a2c2f9056dba79053986f7edbc25b9705"}
{"id": "train_20294", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "reproducibility", "auditability", "governance"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "97e8a630067669175c2a0d047ceb6b9494d8fa27"}
{"id": "train_20295", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "auditability", "documentation", "evaluation_metrics"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "593ad7b4356d14c5e073294b34b0210dba583488"}
{"id": "train_20296", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["auditability", "tooling", "security_gates", "ci_integration"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c41419c5eba8fa123e46e4e6524c739b7fc6a555"}
{"id": "train_20297", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "auditability", "repo_scale_reasoning", "documentation"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "79504825d188add483d82f394c7e313fb86367f9"}
{"id": "train_20298", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "governance", "evaluation_metrics", "tests_are_truth"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fc7771e8ef8e9d8f6309e06df1b82e422593818f"}
{"id": "train_20299", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "tests_are_truth", "security_gates", "auditability"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "28805155a35d2859e80f2f976ac93a62a9cab644"}
{"id": "train_20300", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "documentation", "security_gates"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bdcd57c81c1ce51e544611bf5dff2395f9dce990"}
{"id": "train_20301", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "security_gates", "cost_latency_tradeoffs", "governance"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d457ff1107f2bdb27888fa1b603d8db837b571f4"}
{"id": "train_20302", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "documentation", "reproducibility", "security_gates"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e0f368f5d453f77e729002841344790bac9e2642"}
{"id": "train_20303", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "governance", "documentation", "tests_are_truth"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "78ff4a8de894b647a07cec19e18b6a452daa0f71"}
{"id": "train_20304", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "reproducibility", "tests_are_truth", "ci_integration"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "faf3a099522c3c907843c5c95dbaaa0e33b99f75"}
{"id": "train_20305", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "documentation", "reproducibility"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6f74d976702daacd302300c3223ff18b828ac4d8"}
{"id": "train_20306", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "evaluation_metrics", "tests_are_truth", "auditability"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2ade0c0dcbdc1ee19b9af0e484890f1c3a2399cf"}
{"id": "train_20307", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "governance", "reproducibility", "tooling"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7823b34a9d3b49f2120ba4ee7ad570b23f85f0cc"}
{"id": "train_20308", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "documentation", "evaluation_metrics", "security_gates"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6b7d468dfe32ebe409063c2126423a0d4589f88a"}
{"id": "train_20309", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "documentation", "auditability"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "28cf5859e4c5db33d862cad35d228a1f4efb5d52"}
{"id": "train_20310", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "security_gates", "tests_are_truth", "governance"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7ebbf6e82860f953474251bf0fed4ba3e1be7087"}
{"id": "train_20311", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "repo_scale_reasoning", "ci_integration", "auditability"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ca9fd6d9f890869349b3ac5892a8dc32357aacb4"}
{"id": "train_20312", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "tests_are_truth", "tooling", "governance"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f69973755900739dc807d6e18838751d9c79139e"}
{"id": "train_20313", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "security_gates", "tooling", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "139f04465553fffc3c2878e32ef3ee017a88b7c9"}
{"id": "train_20314", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "evaluation_metrics", "tooling", "security_gates"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d55dd5e875f19e9027d27f054b9333bb985309ef"}
{"id": "train_20315", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "documentation", "auditability", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f04116d361708962a39ae4885c62243b98e8b9e2"}
{"id": "train_20316", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "reproducibility", "repo_scale_reasoning", "auditability"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "56d68f00b629e6d8087b9968114f190060e2e437"}
{"id": "train_20317", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "security_gates", "evaluation_metrics", "governance"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "047690d9feff67c05602c7e7546441bd843ff0ba"}
{"id": "train_20318", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "ci_integration", "repo_scale_reasoning", "evaluation_metrics"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d13acd8f8798f36b41a7db07321d4cd8df269e21"}
{"id": "train_20319", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "security_gates", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bcfdd02fab16d962137080f1da0e6cc53ccf2ff0"}
{"id": "train_20320", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "ci_integration", "tests_are_truth", "auditability"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cf889e54f703cb1900bfb936af02d69338cecb69"}
{"id": "train_20321", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "reproducibility", "ci_integration", "tooling"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "67c6e43020f548ec9e3225f47b69e68678bd1f88"}
{"id": "train_20322", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "documentation", "tests_are_truth"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6219950162418951b7aa8033705cbf8855baf72e"}
{"id": "train_20323", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "governance", "evaluation_metrics", "ci_integration"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "be5383a2f1d51c9d89a05243ff2707154859e8c5"}
{"id": "train_20324", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["governance", "documentation", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0904017b3fdd3ecb90cfa7077a3986bbc274ecf1"}
{"id": "train_20325", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["documentation", "tests_are_truth", "tooling", "ci_integration"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6f35ad4f98ba852cbf6b1203b2b0e02d9197fa33"}
{"id": "train_20326", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "reproducibility", "security_gates", "tooling"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "363b45981050e103e69d912f4959a8979ace0b7a"}
{"id": "train_20327", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "evaluation_metrics", "ci_integration", "documentation"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0c3bbf1e5e9b5ed71de8b1ca461b14165e054787"}
{"id": "train_20328", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "tests_are_truth", "evaluation_metrics", "governance"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5da02fc755075e007214d60e7506e69c632720c5"}
{"id": "train_20329", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "repo_scale_reasoning", "security_gates", "reproducibility"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "13512fa131180ac60930360cdb4a1e7cf4483b42"}
{"id": "train_20330", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "reproducibility", "documentation", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "79e19485a73e135e1e9fb6c265fa115e92b459eb"}
{"id": "train_20331", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "tooling", "governance"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cd40c04fe9b96bfae28456f8213163ea6952e579"}
{"id": "train_20332", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "tests_are_truth", "auditability", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "44ae5cb5e3c4669f0a095a46913905e9a383571e"}
{"id": "train_20333", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "cost_latency_tradeoffs", "auditability", "security_gates"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7f05a3da4f0ecd0be6971bddb5076b89d1a4cc65"}
{"id": "train_20334", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "governance", "cost_latency_tradeoffs", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f85f43a4bd67f27195adb2dda8b7798977fb7d7b"}
{"id": "train_20335", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "governance", "tooling", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "977225192a4094feced41f9cfc22d056516675f3"}
{"id": "train_20336", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["governance", "evaluation_metrics", "tests_are_truth", "security_gates"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f193ba1c4ba36193cc932410cf95751ebda7baad"}
{"id": "train_20337", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "governance", "tooling", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b1f911c120171c05a1dbaf90fae2f86a592f5ea0"}
{"id": "train_20338", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "governance", "repo_scale_reasoning", "reproducibility"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "086bb3fff93504c68f9cc140c7416adeb01562de"}
{"id": "train_20339", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "repo_scale_reasoning", "evaluation_metrics", "tooling"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "09b57087d9277013cdbc595bdd9b089a1cb1fce8"}
{"id": "train_20340", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "auditability", "tooling", "reproducibility"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "71e717c97a756533812f85e5e7e29fa6fc51d146"}
{"id": "train_20341", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "documentation", "tooling", "governance"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e35e0c4b28c3d4c75e683684591f1d71ed31e5b8"}
{"id": "train_20342", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "ci_integration", "security_gates", "repo_scale_reasoning"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0b661fbb0258c3aafb59ae8fd2c5f1c74cee7b1c"}
{"id": "train_20343", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "documentation", "security_gates"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3ee908237be55374a408f95dc99e820436eb6798"}
{"id": "train_20344", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "repo_scale_reasoning", "ci_integration", "evaluation_metrics"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ffcb273c07aa4ad72844e2568031445c8841f8f2"}
{"id": "train_20345", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "ci_integration", "auditability", "evaluation_metrics"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5c469f5469aeb28a4c53e90b0e02a948649aa5b4"}
{"id": "train_20346", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "reproducibility", "governance", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b37c50c499cda72d8844bc3ec5e8ad319ae58872"}
{"id": "train_20347", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "security_gates", "reproducibility", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d269cfc20b84cb668c561c0910895125398a3d83"}
{"id": "train_20348", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "ci_integration", "auditability", "documentation"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "940be5e9d7414ff75da1963ba2e938841f5bfb4b"}
{"id": "train_20349", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "tooling", "documentation"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9ace65a74ee1660664439b9653f32f248d5a26d7"}
{"id": "train_20350", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["documentation", "repo_scale_reasoning", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "805b8154faddbf1963a15a64433bf336d86e4ec7"}
{"id": "train_20351", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "tooling", "repo_scale_reasoning", "auditability"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "790527641a9564b932c048e9e01ecb03044f4bbc"}
{"id": "train_20352", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "security_gates", "documentation", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cb9c50c8b0befef8b26d6c56b6fb7c17c09348db"}
{"id": "train_20353", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "reproducibility", "tests_are_truth", "documentation"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dc18116889b69656cd9abfe4f2acca8f6d454601"}
{"id": "train_20354", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "auditability", "documentation", "governance"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3afafc4c1286e601b3e34833a072d94c285f645a"}
{"id": "train_20355", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "governance", "evaluation_metrics", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a8dc816b5aa8441ce32b29467f3393a80b981faf"}
{"id": "train_20356", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "tooling", "auditability", "repo_scale_reasoning"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4ac14f013f59755f329996b10f1159a6782d35ce"}
{"id": "train_20357", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "auditability", "tests_are_truth"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6a5bd6cf1d6617f59383792d0a49dcd8165e16a5"}
{"id": "train_20358", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "tests_are_truth", "tooling", "documentation"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3f8eaa5883e2f77e0aab8df5f9399103b6ba8f64"}
{"id": "train_20359", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "security_gates", "ci_integration", "documentation"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "57c876e36c1db909a46df5de40acffd1c49c3895"}
{"id": "train_20360", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "568f38e6ce6055300324b1d04db8e289fa9d3b2a"}
{"id": "train_20361", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "auditability", "ci_integration", "reproducibility"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "255a75785e2457f9d645364a0282fc0f446effa5"}
{"id": "train_20362", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["auditability", "repo_scale_reasoning", "evaluation_metrics", "documentation"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "69d796a41004677faedb4fea85996b381fe1eee4"}
{"id": "train_20363", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "governance", "security_gates", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ae5877110e6c9df710ae022e1a460b797ed607f8"}
{"id": "train_20364", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "reproducibility", "repo_scale_reasoning", "security_gates"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e0802ddf35c5890fa5af9db0299178176f9398ed"}
{"id": "train_20365", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "documentation", "governance", "security_gates"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1e4fb45975e3c416a7ca6a5c8dc2f58f0a7f727b"}
{"id": "train_20366", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "governance", "tests_are_truth"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "325a36381ff3c1e098324e0842501abad4fd8bb2"}
{"id": "train_20367", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "auditability", "documentation", "security_gates"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2f85335ab5fc955790e3dd944a5c28431af0434d"}
{"id": "train_20368", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["auditability", "documentation", "tooling", "reproducibility"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7e3a105a185cb4e9c30cb69af33feb6d3587c054"}
{"id": "train_20369", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "reproducibility", "tooling", "ci_integration"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1f6500f013165e6d12fed3bc591887c9ba95c76a"}
{"id": "train_20370", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "tooling", "evaluation_metrics", "ci_integration"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3f6d145acf8a931d4cca556d8ed13451903d80c5"}
{"id": "train_20371", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "tests_are_truth", "documentation"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ee5a26d7815d2efa851f89c3692b4cbd2af94720"}
{"id": "train_20372", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "tooling", "documentation", "auditability"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2a7d117f046270288f416902d4cb333f0f7fe9ea"}
{"id": "train_20373", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "tests_are_truth", "auditability"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2d7dd1021a3535f88be465a5960081ecfc8f5da5"}
{"id": "train_20374", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "reproducibility", "auditability", "tests_are_truth"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9965357af191ff91c29ec8fcfc509c711f726825"}
{"id": "train_20375", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["governance", "cost_latency_tradeoffs", "security_gates", "tooling"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a9e1a200750a83f26cb7b6670c52163e0171461c"}
{"id": "train_20376", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "reproducibility", "documentation", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "54031425e0e64b261b2d535a77ab6bd15b6a7473"}
{"id": "train_20377", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "tests_are_truth", "repo_scale_reasoning", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "42e963f016de6732bb45009e463d598cf66389ce"}
{"id": "train_20378", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "tooling", "documentation", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "73fce57081cdf2ffdc0ea156ab5756fa608b133f"}
{"id": "train_20379", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "tooling", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2b6463910e3a799b6745dfa626ef4580d8a64df4"}
{"id": "train_20380", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["governance", "reproducibility", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0838aec18c26f710b0fd391fd838a190ff6a2259"}
{"id": "train_20381", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "evaluation_metrics", "governance", "documentation"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e609480a1c7f2ac2fb74d6deef21a664a5b05610"}
{"id": "train_20382", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "auditability", "governance", "documentation"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "25f88e1a689371d20e1af11e44e7e697a7594838"}
{"id": "train_20383", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "security_gates", "auditability", "evaluation_metrics"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4704dd4a634320770028271a0237ea0634f913d5"}
{"id": "train_20384", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "documentation", "security_gates", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "73525d9bdef29473bc64d803ea8d37f04976bb69"}
{"id": "train_20385", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "ci_integration", "tooling", "tests_are_truth"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dab13c4b26a1b99d0747a19cdf0943a59e974cae"}
{"id": "train_20386", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "security_gates", "tooling", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "290eae7b5df7d09532ead4a5eb897fd95bc3b7ec"}
{"id": "train_20387", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "auditability", "tests_are_truth", "governance"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b243227be1d35665108498112b1dd023849ca01b"}
{"id": "train_20388", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "governance", "security_gates", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7eb3470951ee9df997e1825a8ec9bbbbbbbc6cb0"}
{"id": "train_20389", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "ci_integration", "governance", "auditability"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5cacee1c5ebbb9be660b7fb96e6d5c14adaf382d"}
{"id": "train_20390", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "auditability", "evaluation_metrics"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7e5e1353ad147f949bd5f0db92046829aecbb99b"}
{"id": "train_20391", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "tests_are_truth", "tooling", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3886d89a7577f0f18de85b77c17739b444fe8f0b"}
{"id": "train_20392", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "tests_are_truth", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5bf3dd125247e36e7d06813cdd0004153083779e"}
{"id": "train_20393", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "tests_are_truth", "reproducibility"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cfdccd36f1ce483f05dd490788b0707a406e51a5"}
{"id": "train_20394", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "documentation", "auditability", "tooling"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1aafe2e41257e3aa937d91f6954b8187daad4276"}
{"id": "train_20395", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "ci_integration", "security_gates", "repo_scale_reasoning"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4b90a026cf8ba1906c75458c431006edfe231c17"}
{"id": "train_20396", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "governance", "documentation", "tooling"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4257558c961ed704236fe21e638e8cd4b6e75368"}
{"id": "train_20397", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "reproducibility", "ci_integration"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1c44fb00b2fd77c2020b7097fcc440a310956456"}
{"id": "train_20398", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "auditability", "tooling"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5ee0d4e6e41056bf6101881a21759fba51ad156d"}
{"id": "train_20399", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "evaluation_metrics", "documentation", "auditability"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d16f068cd5f44e0520eb3bdc35e8a686bc37a6f0"}
{"id": "train_20400", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "evaluation_metrics", "reproducibility", "documentation"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cb9c50c8b0befef8b26d6c56b6fb7c17c09348db"}
{"id": "train_20401", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "auditability", "tooling", "documentation"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0071a466c65c27287f837cd384f504d836564603"}
{"id": "train_20402", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "reproducibility", "auditability", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "690b0a4b256afeaa8a702384a50eb075e7d0292a"}
{"id": "train_20403", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["governance", "cost_latency_tradeoffs", "tests_are_truth", "tooling"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "702853629ad2459f546e6d14cd212148d1169f54"}
{"id": "train_20404", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "tooling", "reproducibility", "governance"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1a89be84e8d6e6b59c5546c9fa97bf7797ef64de"}
{"id": "train_20405", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "governance", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "54c29285c39f382332168f30ebeec0737985157d"}
{"id": "train_20406", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "auditability", "ci_integration", "tests_are_truth"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ad47d8d7cbb6a574c5a574fd52d007bfb2b63ebe"}
{"id": "train_20407", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "reproducibility", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "61fce0b2cabf47c28c49ff20bcfb05e9e1947c1f"}
{"id": "train_20408", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "security_gates", "evaluation_metrics", "tooling"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ac765edb38930627533bc365a641f85cdc113606"}
{"id": "train_20409", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "tooling", "governance"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6ed0551779f5236100d8c0e05128699eae668f8b"}
{"id": "train_20410", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["governance", "cost_latency_tradeoffs", "reproducibility", "ci_integration"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "493d0d5dafb54380d9d408b2f1b731dff727e4fd"}
{"id": "train_20411", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "reproducibility", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b4ee26dcdba58ad8cbafbfb293ad1f6ac34f10a0"}
{"id": "train_20412", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "reproducibility", "repo_scale_reasoning", "security_gates"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fdc01abb72e802c7a7743ebd3e954f45965bd6da"}
{"id": "train_20413", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "documentation", "security_gates", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1810842f2863aa6374f62d33b74fbc69216a6a5c"}
{"id": "train_20414", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "ci_integration", "tooling", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a10bebd902ff4d53b404ab707709f896dcdd8745"}
{"id": "train_20415", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "auditability", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6410f3fc10c933a8bc309bf27023368ad7456dd6"}
{"id": "train_20416", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["governance", "repo_scale_reasoning", "documentation", "security_gates"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d722b93bbfe7c2d8cb5914a96508b5cbd11c1a47"}
{"id": "train_20417", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "reproducibility", "ci_integration"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "927333f72c74d5389ab6d35ca85a21beaaecfa52"}
{"id": "train_20418", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "ci_integration", "auditability", "tests_are_truth"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "47a3095756bdeb6a011802700ca47d4258d78ff0"}
{"id": "train_20419", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "repo_scale_reasoning", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "055e709dcfb8fbc3e951fb820f813433bdd60276"}
{"id": "train_20420", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "tooling", "auditability", "security_gates"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bc6e8aa02078f6b9b36b095406f43dd8dd619cf5"}
{"id": "train_20421", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "cost_latency_tradeoffs", "governance", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "483800a9d253a3713e5029f3abe793afc27a21f9"}
{"id": "train_20422", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "ci_integration", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "075c213261dbc19c574b04d5b063c08830d6f19b"}
{"id": "train_20423", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "governance", "evaluation_metrics", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "00d01ba7d54f406f53131741d61bded26c6f4212"}
{"id": "train_20424", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "reproducibility", "ci_integration", "tooling"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "284ab27e7f3883f53170f1c6d3d4753a46afca58"}
{"id": "train_20425", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "tooling", "governance", "repo_scale_reasoning"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "95b1435a52b0cc9e67de0a4f6ad5f03e2bd76539"}
{"id": "train_20426", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "documentation", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f9c43397811f277df0c933b5b50c4742de3191fb"}
{"id": "train_20427", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "tooling", "ci_integration", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "269d89dbdfd2c234984f1967c9f8027eeea1e0a2"}
{"id": "train_20428", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "tooling", "reproducibility", "ci_integration"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "eb01fe8d083f561e79c322329f3e1df7006dc309"}
{"id": "train_20429", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "governance", "security_gates"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9662b1bf8d43fdae7a22c36b296f6cac67f776bb"}
{"id": "train_20430", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "documentation", "tooling"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0d9da5af97d26d8183b087e4e7a94a7e7954fc3f"}
{"id": "train_20431", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "auditability", "governance", "ci_integration"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "679044c575fbcb6e4b462cb5919d4230cd61153a"}
{"id": "train_20432", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["auditability", "documentation", "cost_latency_tradeoffs", "security_gates"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3d09cec57ae79c362a9986882402146d1858b5cc"}
{"id": "train_20433", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "tooling", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8c6d0f24f888c9d234d5fb392579c0d0c9df2548"}
{"id": "train_20434", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "ci_integration", "auditability"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "336f482152db4aa2a53b34aed475a6632f5a542c"}
{"id": "train_20435", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "documentation", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5d54a6ca7b3906cd37184ce18d2f6faabc316aa1"}
{"id": "train_20436", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "auditability", "ci_integration", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3af89e68894b63a4f714a9df6d27e8dee8bca8ea"}
{"id": "train_20437", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "security_gates", "documentation"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0d1eee774ae9630d0579b0d6fd872f145e6c97de"}
{"id": "train_20438", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "security_gates", "auditability", "tooling"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "07a35cb0a6d9b7882d37ed7a76be4bcdbc8f955c"}
{"id": "train_20439", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "tooling", "governance"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a5dbda55964bf749a21fb040a8ffef5d5310b57c"}
{"id": "train_20440", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "documentation", "auditability", "reproducibility"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "492c12e8a64e2de8ea4c56ced652485430c01f9f"}
{"id": "train_20441", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "governance", "ci_integration", "evaluation_metrics"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "802431164c1859b38eb8b1c9ae1723545a2c638e"}
{"id": "train_20442", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "reproducibility", "tests_are_truth", "governance"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0d401012124d24e946a14b6dcf999b811c2bbe55"}
{"id": "train_20443", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3c94384a05b578e128da8b4f181255539fac94ae"}
{"id": "train_20444", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "ci_integration", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "701c25faed46b84d25846bc0cc03719736d1ddf8"}
{"id": "train_20445", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "evaluation_metrics", "tests_are_truth", "auditability"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fea0bd741622a3dff2cb38856dc1b49232f37001"}
{"id": "train_20446", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "governance", "security_gates", "reproducibility"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c7e957bc76cf45f231fe4246e3d5622ebff12ed7"}
{"id": "train_20447", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "repo_scale_reasoning", "reproducibility", "tooling"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "55a869b6853401241f18c95dc0fe0703306da9d5"}
{"id": "train_20448", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "tests_are_truth", "ci_integration", "governance"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5b1d32f915b6a4b9ea124b923bbd3d4c841a1931"}
{"id": "train_20449", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "ci_integration", "evaluation_metrics", "tooling"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bc6e8aa02078f6b9b36b095406f43dd8dd619cf5"}
{"id": "train_20450", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "ci_integration", "tooling", "auditability"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f34eca0edb6e097298c19057d0ea74d54a8bd0d5"}
{"id": "train_20451", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "tooling", "evaluation_metrics", "auditability"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "23287131987a6d30a5e8119318d81045a4e749d6"}
{"id": "train_20452", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "tests_are_truth", "repo_scale_reasoning", "security_gates"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "36eb0e8b933994d35ca940822ea21b16e7a31949"}
{"id": "train_20453", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "tests_are_truth", "security_gates", "auditability"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d4ae3317f4b12c9082921bdd9e73b639435f84d3"}
{"id": "train_20454", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "auditability", "reproducibility", "tooling"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2f2bc0f28750c16f0e6c92bbdfda334e84bbab8f"}
{"id": "train_20455", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "tests_are_truth", "tooling"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "37b8f483d07cbb208a862388ad22e7aad2843f92"}
{"id": "train_20456", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["governance", "evaluation_metrics", "documentation", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4c178888eecf7dfe4c314afca4ef8f79d3fc533a"}
{"id": "train_20457", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "tooling", "governance", "tests_are_truth"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1e84a82ccc881ad868fbcb728e541c933ed1198b"}
{"id": "train_20458", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "reproducibility", "cost_latency_tradeoffs", "governance"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "948ed6d97dac029462b2262d0e36729d50581254"}
{"id": "train_20459", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "governance", "tooling", "reproducibility"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dbc782db23ef8469d7c803b5e75ec66279ac015e"}
{"id": "train_20460", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "documentation", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a870d79cb12c5888e0f3c43f8e6f057caa9b683a"}
{"id": "train_20461", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "auditability", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dc42d6ccda679bccdfe3c7f03869dc85da2f55a9"}
{"id": "train_20462", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "governance", "tooling", "security_gates"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "641df3573b0ccddfdb8f29c2285e28af48dfba5a"}
{"id": "train_20463", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "evaluation_metrics", "security_gates", "tests_are_truth"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fb1c319751c2872da4bfc5782b732f179ee9410b"}
{"id": "train_20464", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "tooling", "governance", "ci_integration"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "142d7f60d1fdca4fe30010e78069e6b39ee21cf7"}
{"id": "train_20465", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "cost_latency_tradeoffs", "evaluation_metrics", "ci_integration"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d392c5ccb74c10e0b14a0384335fe81875ffc61b"}
{"id": "train_20466", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "repo_scale_reasoning", "tests_are_truth", "tooling"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "28061a40938ed771679b42d699ff14c91157fbc6"}
{"id": "train_20467", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "tooling", "documentation", "auditability"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2fc6bd0d72794a4f472ae5781a093a9f49bf0e0f"}
{"id": "train_20468", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "tests_are_truth", "tooling", "auditability"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "16dd92e4930822565877b171b72da9567a310e2b"}
{"id": "train_20469", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "auditability", "reproducibility", "governance"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e6b044c188e8f2bc73eca523261ef42b2f358374"}
{"id": "train_20470", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["documentation", "cost_latency_tradeoffs", "governance", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "21c45c852bd1d5648611c00ce0929ec1644338e1"}
{"id": "train_20471", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "tooling", "reproducibility", "auditability"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6913958e32c733bbca22e9f323071ce72c588674"}
{"id": "train_20472", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "security_gates", "documentation"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "59ee8b87d26e6796072404a0625d87a0aae38c7c"}
{"id": "train_20473", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "ci_integration", "documentation"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1e4fb9ad7499197706424189cdc0a43435fbf876"}
{"id": "train_20474", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "tooling", "documentation", "governance"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b7644151b16a1a42ff7aa44021451dcc0e292f2c"}
{"id": "train_20475", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "security_gates", "governance"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e17714401e583a8b0ca31a0816d5a536acff615d"}
{"id": "train_20476", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "tooling", "reproducibility", "documentation"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ec0df10f85a17dd97c4f94f81ea232a04cc2b1f2"}
{"id": "train_20477", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "evaluation_metrics", "auditability"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a4dd36662148c54c4e3d0f4b8ee3789e60187b72"}
{"id": "train_20478", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "security_gates", "evaluation_metrics"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3dc6f8ec496ecc533992d48da79d1d7af80aff56"}
{"id": "train_20479", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "cost_latency_tradeoffs", "repo_scale_reasoning", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ab4d4e9983bdd980f4b56b6e2785f9beb8a772c5"}
{"id": "train_20480", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "tests_are_truth", "cost_latency_tradeoffs", "governance"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "64e4571618e0986c4dbbbc799e1e676739bbadaa"}
{"id": "train_20481", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "auditability", "evaluation_metrics", "ci_integration"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f53710748ed8858991c1bab159ad1119962d4f18"}
{"id": "train_20482", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "governance", "ci_integration", "security_gates"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ab7ec0c4693dca91677172b877054c9cff598e2e"}
{"id": "train_20483", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "reproducibility", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "09f35a6115a96f8824c8998528e9b4a6402c2611"}
{"id": "train_20484", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "ci_integration", "governance", "documentation"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d2fb2af9b00b0e74b6a37db50c69db42a34d1e0d"}
{"id": "train_20485", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["documentation", "security_gates", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8e84758ee125f7a3fb40781060b064a698dfb545"}
{"id": "train_20486", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "tests_are_truth", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d7fe4f2db975bafad3673d44020efe411eb51c55"}
{"id": "train_20487", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "tooling", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "982bbcb56448c198d74919084d9174c8b443c84a"}
{"id": "train_20488", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "tests_are_truth", "documentation", "auditability"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8c7b5e1598ccdb34789d3086fc35c1fc2b0de059"}
{"id": "train_20489", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "security_gates", "documentation", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b8fca6aa60b1eda617d5ecc68da721d19d4fec6b"}
{"id": "train_20490", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "ci_integration", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0c46bfa101d332b05ecf7c55020102d00671cba0"}
{"id": "train_20491", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "auditability", "reproducibility", "evaluation_metrics"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "16e8873979faad7a819eb50477afb102336805b8"}
{"id": "train_20492", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "tests_are_truth", "ci_integration", "evaluation_metrics"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "20853dfd309e63349d481751e9a6d8cee7a1435e"}
{"id": "train_20493", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "governance", "tooling", "evaluation_metrics"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d247d308407e4b271b018f3e97562128f69bf7bd"}
{"id": "train_20494", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "documentation", "cost_latency_tradeoffs", "security_gates"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "aabe3cbddfd28160c37bd262e6cc9009382dd16c"}
{"id": "train_20495", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "security_gates", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ddd0712ec8bf14975db9ca24da83d47ce1595284"}
{"id": "train_20496", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "security_gates", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d27aebe97df48fe4ef3a6f0c777e405b064f9f99"}
{"id": "train_20497", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "repo_scale_reasoning", "ci_integration", "tooling"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "78257474afc370960e49641b687cd19ce2bc5410"}
{"id": "train_20498", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "evaluation_metrics", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b9db2a6d67881bd38698c69a9a3353b0c7835361"}
{"id": "train_20499", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "security_gates", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8f1550602e9a44fc9eaf4b3b203b6781aa17a2b2"}
{"id": "train_20500", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "documentation", "reproducibility", "auditability"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1d6304a6f536583ab57b342a59cf5a2f2feac519"}
{"id": "train_20501", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["governance", "reproducibility", "ci_integration", "evaluation_metrics"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8747185a7f5d61f21905c1e239542135ba6ed9b2"}
{"id": "train_20502", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["governance", "tests_are_truth", "security_gates", "ci_integration"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "592d6f445a2992b62f4dfe0d80a5f603b2c3f0af"}
{"id": "train_20503", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "reproducibility", "security_gates", "documentation"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "99542cca70308c9466dc68c7c2a2165295f93cef"}
{"id": "train_20504", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "auditability", "tooling", "ci_integration"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "032a256a75643defc64deae34f553ef5f9c78c81"}
{"id": "train_20505", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["auditability", "documentation", "governance", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d3c5d0603bcd2cf1a93edbbca40983e6373e46c0"}
{"id": "train_20506", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "tests_are_truth", "governance", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bd1298263100be33a11ad8f620bf921534f450a1"}
{"id": "train_20507", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "governance", "ci_integration"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a4a801b4ab6bad7db682198fc2f72e0d206cc3e3"}
{"id": "train_20508", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "governance", "evaluation_metrics", "tests_are_truth"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "56373181aced1235c0ef94b3449072131791a646"}
{"id": "train_20509", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "documentation", "tooling", "tests_are_truth"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3c281bc52da1c0a3eaa5173dd4926911aee906c6"}
{"id": "train_20510", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "tooling", "ci_integration"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8da6080d73f1970c545a5968b4f6663c3b09c8df"}
{"id": "train_20511", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "tests_are_truth", "cost_latency_tradeoffs", "security_gates"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2017a84b0817d380337406832f1522fd7af3fef4"}
{"id": "train_20512", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "security_gates", "repo_scale_reasoning", "reproducibility"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "35744582dd22a7ae645f23bb898b43168904c59d"}
{"id": "train_20513", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "tests_are_truth", "repo_scale_reasoning", "governance"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "07ed27097fba57a0f15f7c1c572bba75602b0621"}
{"id": "train_20514", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "tooling", "evaluation_metrics", "documentation"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bebd13c45819a05f2be28d6ac219a8711e100bfd"}
{"id": "train_20515", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "evaluation_metrics", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cd2ca39e588e9a7aa27fc2268acf6692963de60c"}
{"id": "train_20516", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "tests_are_truth", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0048dff59bf428d4794e56dc8af6abd9ecd84bbb"}
{"id": "train_20517", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "ci_integration", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c4d5c38e7f7ecb7b9131416143da8bb3911a3eb7"}
{"id": "train_20518", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "repo_scale_reasoning", "tests_are_truth", "ci_integration"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bcb258de2e6cd99b05540b06426cdc6758b37962"}
{"id": "train_20519", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "documentation", "ci_integration"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4b8d76f20eb0da2cc4b8654038ef00ef90035ad3"}
{"id": "train_20520", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "governance", "auditability", "documentation"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5249582aa425cc5361a0d64a691685986b7482f8"}
{"id": "train_20521", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "security_gates", "tooling"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "df02baebe799ba92785a791e91f702dcc906f153"}
{"id": "train_20522", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "auditability", "evaluation_metrics", "tooling"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9a6f563302d8c9f283641e3393542d493af44787"}
{"id": "train_20523", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["governance", "ci_integration", "auditability", "security_gates"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "35bf0bebae0d019e7c864519ea7666d5cdaa3d80"}
{"id": "train_20524", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "ci_integration", "reproducibility"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "37b8f483d07cbb208a862388ad22e7aad2843f92"}
{"id": "train_20525", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "documentation", "tests_are_truth", "ci_integration"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a6ec48c4c10b8be47d6165e45830487e607dce36"}
{"id": "train_20526", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "reproducibility", "tooling"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ced72b525d2ee31b3d8e3c3e1f66182868394830"}
{"id": "train_20527", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "tooling", "repo_scale_reasoning", "documentation"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "26d89b049c843d1cd0504067a60c2345706ee631"}
{"id": "train_20528", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "tests_are_truth", "tooling", "reproducibility"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "73e50e1922562243093d2f2a8a1da93358741e6c"}
{"id": "train_20529", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "tooling", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "56248a926f242e69e16ebe6e99fd3f415c58e809"}
{"id": "train_20530", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["governance", "repo_scale_reasoning", "evaluation_metrics", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "41be198fc47f5e1fe699f437bc65846a7ab19257"}
{"id": "train_20531", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "governance", "repo_scale_reasoning", "auditability"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4471b0d4dd0468312fb4699270824b1e983f6597"}
{"id": "train_20532", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "auditability", "evaluation_metrics", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c2568660b6130d75409e3a9e5d68d87442e80d2f"}
{"id": "train_20533", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "evaluation_metrics", "security_gates", "governance"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f446f2b6dbda273e139dbca0b903802e43227702"}
{"id": "train_20534", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["documentation", "repo_scale_reasoning", "tests_are_truth", "tooling"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0c1451c9d1e91b34cec658e06d3c383046f91736"}
{"id": "train_20535", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "reproducibility", "ci_integration", "tests_are_truth"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a2ed8ed2b73b2b25ea96cb8403db9558a049f5fc"}
{"id": "train_20536", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "security_gates", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7893549bd34fba90e7b69930aa625538258747e6"}
{"id": "train_20537", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "reproducibility", "security_gates"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cc7379f5cd2f72fc52ac32c8a70de6b1811c8edf"}
{"id": "train_20538", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "repo_scale_reasoning", "security_gates", "tests_are_truth"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d056ea3e21c528c405fc3161203d20642c38705a"}
{"id": "train_20539", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "evaluation_metrics", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fad8c3298a21820a649afa41e011bfbfda5741f8"}
{"id": "train_20540", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "documentation", "ci_integration", "security_gates"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "27677e025a29ab9aebd616f4107d57e4ccd5202b"}
{"id": "train_20541", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "repo_scale_reasoning", "tooling"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0b10b721e50b21c67bab2807f0651a58697a8e73"}
{"id": "train_20542", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "evaluation_metrics", "security_gates", "ci_integration"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1e88ea99e21fdd479dba6d1f1d41827c03347743"}
{"id": "train_20543", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "auditability", "ci_integration"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dc632641d46e21aee611014b190f658c816f94db"}
{"id": "train_20544", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "evaluation_metrics", "cost_latency_tradeoffs", "repo_scale_reasoning"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "10bfdd0209646e15addd52017216d2031c61ee12"}
{"id": "train_20545", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "tests_are_truth", "reproducibility"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c24f8de6cb7ee029952159ab4d5df832565435ca"}
{"id": "train_20546", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "reproducibility", "tests_are_truth", "security_gates"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4bf8e5890fcbdbc1f63d22681c287961708bc490"}
{"id": "train_20547", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "tests_are_truth", "ci_integration"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e531473fbb3c86823b6382cf519760019c14735a"}
{"id": "train_20548", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "documentation", "tooling", "repo_scale_reasoning"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d462df37f458ac69dfe412eacf154b771467ffea"}
{"id": "train_20549", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "repo_scale_reasoning", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "597fcc6d36c057be856520f0a5641f6f59122350"}
{"id": "train_20550", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["governance", "security_gates", "auditability", "tooling"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ddd7e0a18bd0f7c8a687a17f23ab8fd533cf96cc"}
{"id": "train_20551", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "governance", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0eb8a06942b0211868595ddc62fff87bc210d941"}
{"id": "train_20552", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["governance", "evaluation_metrics", "auditability", "tooling"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5032ef50ba56b1f10458f3cfad34f49964ce708d"}
{"id": "train_20553", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "governance", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "253a05b716a12460724a5f0d69076cfc7c62989d"}
{"id": "train_20554", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "ci_integration", "documentation"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4fe2e0d1985177a8034fbfde2af340c426650f95"}
{"id": "train_20555", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "reproducibility", "governance", "auditability"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4f496a7a88e6055717f6aed9b2cbf4b864a3d9e6"}
{"id": "train_20556", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "security_gates", "documentation"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "00280af5f7587b822daefb4a9ca44b028e273964"}
{"id": "train_20557", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "tests_are_truth", "reproducibility", "security_gates"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6bc9ed12ff71c364809defe21a623a16abef67c9"}
{"id": "train_20558", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "documentation", "governance"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f4d97396d7761618968652ff2514f2e77a331e1e"}
{"id": "train_20559", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "auditability", "tooling", "ci_integration"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "afa9982da51ccad72c0aeee14dfd0e0b921ec562"}
{"id": "train_20560", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "security_gates", "auditability", "ci_integration"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "46a12ed24b3d068ea01a4b2575307dbc4e0bee26"}
{"id": "train_20561", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "reproducibility", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7a6b8914267fe52193edbfdfb78034884c2d46ab"}
{"id": "train_20562", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "tooling", "evaluation_metrics", "reproducibility"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d8dff947cd3c46b020f9d712cf5724ffae43594d"}
{"id": "train_20563", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "tooling", "ci_integration", "tests_are_truth"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ddd02f85e5bcd42bf2e3dea1fcf6e4a9c60cf44d"}
{"id": "train_20564", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "tests_are_truth", "governance", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "54031425e0e64b261b2d535a77ab6bd15b6a7473"}
{"id": "train_20565", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "tooling", "tests_are_truth", "reproducibility"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cdcf32568b9063eae2d70b3e814d01c792423866"}
{"id": "train_20566", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "governance", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d5c2be0c741cbff3fa462a8d39ab8bc1e1a24379"}
{"id": "train_20567", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "repo_scale_reasoning", "security_gates", "tests_are_truth"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "20e14ab0532a12f0ea4e37022a239f613e185dcb"}
{"id": "train_20568", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "governance", "ci_integration", "auditability"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cf31a00501e73e7f74c47028c42c4e06689b540b"}
{"id": "train_20569", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "tooling", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1a89be84e8d6e6b59c5546c9fa97bf7797ef64de"}
{"id": "train_20570", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "documentation", "governance", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cfdd54816012e1c1f90c27ea3d217fdd2b707ec9"}
{"id": "train_20571", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["auditability", "tests_are_truth", "reproducibility", "tooling"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "866254a7dac49c99f9fe7da7d7b66ff5666f84e3"}
{"id": "train_20572", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "security_gates", "tests_are_truth", "documentation"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "843ea2b61043a0d10b6cfe8ba2048a787fe3b0fe"}
{"id": "train_20573", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["documentation", "reproducibility", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9e7ac289ee1cfb5830ae554724a17235da2ca5a2"}
{"id": "train_20574", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "evaluation_metrics", "auditability", "tests_are_truth"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8a72187bef4a32e9da7b609593101ce7719f3275"}
{"id": "train_20575", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "auditability", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "af64b903e8476d55e7e80a32df13915237b7eff0"}
{"id": "train_20576", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "auditability", "ci_integration", "tests_are_truth"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f46003ac356a092d66fd1d51718d1660974fca2d"}
{"id": "train_20577", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "tests_are_truth", "reproducibility", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ea1cf397984a67b7e8c19b38836236fb0831b6b5"}
{"id": "train_20578", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "documentation", "tests_are_truth", "security_gates"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bc07fe33f05cf1b570cc09ee1d3a057a0747fad7"}
{"id": "train_20579", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["governance", "security_gates", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0bf037373edfaadeeb9bd53cb672cb1c10945f36"}
{"id": "train_20580", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "tooling", "security_gates", "tests_are_truth"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5b8e452fc9dd21fd161966be49c899f304a1bfa8"}
{"id": "train_20581", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "documentation", "evaluation_metrics"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "924c0507fefd78e67bbeb637fe30641f2901735e"}
{"id": "train_20582", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "security_gates", "documentation", "tests_are_truth"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3e546e8f50b831d71b6321a895435702452dcaf4"}
{"id": "train_20583", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "governance", "evaluation_metrics", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f222e54c10d7c498cfc42817ed0bfd8f594989bb"}
{"id": "train_20584", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["governance", "reproducibility", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "566099a6332c3b6d6cadde6a591239b1069c5b2a"}
{"id": "train_20585", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "governance", "reproducibility", "evaluation_metrics"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7f9e69d517e00474f1fcf0c040989afacee853c1"}
{"id": "train_20586", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "auditability", "security_gates", "governance"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3d2f4b30364cf14638e3a7a163be5ab815329340"}
{"id": "train_20587", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "tooling", "tests_are_truth", "governance"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "46dc563a87ae6761b986a82142c304fca2157be2"}
{"id": "train_20588", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "ci_integration", "repo_scale_reasoning", "security_gates"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bc6e8aa02078f6b9b36b095406f43dd8dd619cf5"}
{"id": "train_20589", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "tests_are_truth", "auditability", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "82eab93b9b93242c0b57bff4e6e528cb0d6c5efb"}
{"id": "train_20590", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "ci_integration", "tooling"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "303bf4a7626689769aabe14f8a03ddf5a8d6da2d"}
{"id": "train_20591", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "reproducibility", "documentation", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "71d3d247c50c9d6e99450d3136245c1347eabe41"}
{"id": "train_20592", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "evaluation_metrics", "tooling", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ecbeb5a98493f702f219e9778ce11b36185969d4"}
{"id": "train_20593", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "security_gates", "documentation", "auditability"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "caf01ef2dfe51ca537116a188d195c9c96bf09ee"}
{"id": "train_20594", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "documentation", "tests_are_truth", "auditability"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "882f83b546a7264dafd9b26949c5ff6070f072ec"}
{"id": "train_20595", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "evaluation_metrics", "auditability", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9d332fe69c3d7f37cae2d577b1eec002279891b7"}
{"id": "train_20596", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "security_gates", "reproducibility", "ci_integration"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "629dc048a2b99cb15142a4941b71bf8d0fc3c279"}
{"id": "train_20597", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "documentation", "ci_integration", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "273810dda51ab2122b3bc7b12c8835c0066c22a0"}
{"id": "train_20598", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "tests_are_truth", "cost_latency_tradeoffs", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ca5ac0582c1aa918dad80fcbda56bbbf850069a9"}
{"id": "train_20599", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "ci_integration", "evaluation_metrics", "repo_scale_reasoning"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c592d178d54f58e2d00381a7385094df7e101209"}
{"id": "train_20600", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["governance", "tests_are_truth", "security_gates", "auditability"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3cd192947b6c36be018fc4c47fff1ee42f1fedb7"}
{"id": "train_20601", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "ci_integration", "tests_are_truth"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a2bb68bb34b6749ffdd4922fa2d6c577d4f13d27"}
{"id": "train_20602", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "repo_scale_reasoning", "governance", "security_gates"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b60fc2e6ad024f9d15473a8cc1e4e96b60a167ec"}
{"id": "train_20603", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "evaluation_metrics", "auditability"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2be88a2f9a411c24e21c9b8e392699eabfef58d3"}
{"id": "train_20604", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "reproducibility", "auditability"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "895b1e1cc947347695d8e694ba91181d79630187"}
{"id": "train_20605", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "documentation", "reproducibility", "auditability"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4b0ceac01e58c0256c838a02836dd1709789e418"}
{"id": "train_20606", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "auditability", "tests_are_truth", "reproducibility"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "37f2865bbcb42521b07506917e3e8edef2692e82"}
{"id": "train_20607", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "documentation", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d5a844e5afc24210f7b4a0ddeb1fc7a7b2546e12"}
{"id": "train_20608", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "auditability", "documentation", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e3fceb7e50920614a490af6cb8237f58ec589ee7"}
{"id": "train_20609", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "ci_integration", "reproducibility", "governance"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d5827b91808c6e2428f1937cefe7d42b0b663656"}
{"id": "train_20610", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "tests_are_truth", "security_gates", "tooling"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f81bced08e39406940dcf5c9bdd54027aba02097"}
{"id": "train_20611", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "documentation", "tooling", "reproducibility"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "29e082738a49ce02e0fa1a123fa75329fe28b5d1"}
{"id": "train_20612", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "reproducibility", "security_gates", "tests_are_truth"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "92a04c10c7a1c2f551b2d118c2ff858e20353af1"}
{"id": "train_20613", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "evaluation_metrics", "ci_integration", "security_gates"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bcad9d5f7885ecf03961aad75ec0e9a84301bac8"}
{"id": "train_20614", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "auditability", "ci_integration"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ccbe964c54f0b5722208fc05362631c265a248c6"}
{"id": "train_20615", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "tooling", "auditability", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9511e8cd50a947ac8c2a2774e748c913e4e908ad"}
{"id": "train_20616", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "documentation", "security_gates", "ci_integration"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2fbd1490babd4386083f7566254084450c291b94"}
{"id": "train_20617", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "tests_are_truth", "security_gates", "documentation"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "90f714ddca20ddf275703e3073d223437e1e62f0"}
{"id": "train_20618", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "reproducibility", "security_gates"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f394b9bfb36f5ec399857b242ab10bf5ae689c62"}
{"id": "train_20619", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "auditability", "ci_integration", "tests_are_truth"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c6137da2e68581f48125c01cd8fbfbfda88752f2"}
{"id": "train_20620", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "reproducibility", "documentation", "ci_integration"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0ac9d6999840cdb8730d488add3d07b4e20eafa9"}
{"id": "train_20621", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "documentation", "reproducibility"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "589cea79f06b30bf029e112cb4d7d27d77c646c4"}
{"id": "train_20622", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "reproducibility", "auditability", "governance"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "88fdaaf93464954b63cf242c06f3c5ad0c0fde29"}
{"id": "train_20623", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "cost_latency_tradeoffs", "tests_are_truth", "ci_integration"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "21ece463daf00ff0d2dbd1f155e9b0a2931ffc15"}
{"id": "train_20624", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "governance", "security_gates", "evaluation_metrics"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "376916fee74317ad9bbdcb48153fad5df3314f31"}
{"id": "train_20625", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "evaluation_metrics", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0118ade6aa42a4a569832f1426720bd1f8672ca5"}
{"id": "train_20626", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["auditability", "ci_integration", "tooling", "reproducibility"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9500c717c13236f2837f2cadaf17cbd26b01359d"}
{"id": "train_20627", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "documentation", "governance", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e6309203e240aec3a3648bdb0f9a5f44fe7d3cf0"}
{"id": "train_20628", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "documentation", "governance", "auditability"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "12529f56183b1833ac9d9d2dc46921d7791f9aab"}
{"id": "train_20629", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "ci_integration", "security_gates", "reproducibility"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "644f43bfa5f169b775a6d7e565a63bbaeaec811e"}
{"id": "train_20630", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "evaluation_metrics", "auditability", "tests_are_truth"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5bf3dd125247e36e7d06813cdd0004153083779e"}
{"id": "train_20631", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "reproducibility", "tooling", "tests_are_truth"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "858b1420c2749dcdf38669ade65c405cfadc48a1"}
{"id": "train_20632", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "cost_latency_tradeoffs", "security_gates", "governance"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "012e1e44c0a5a1781356f4e7a664400b2ed6ce04"}
{"id": "train_20633", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "auditability", "tooling", "security_gates"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "75abd1783603dcc42ccf02fa393b0215a5a5b2ce"}
{"id": "train_20634", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "repo_scale_reasoning", "evaluation_metrics", "reproducibility"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ded5c643a9d74d88f7eac911892c28d7a41fae50"}
{"id": "train_20635", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "tests_are_truth", "documentation", "reproducibility"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "455c9e01cec4ee6d408fb456874437484cbc0ee5"}
{"id": "train_20636", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["auditability", "governance", "documentation", "evaluation_metrics"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8788882e8ac6191e058411bf940ea946839fdc92"}
{"id": "train_20637", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "evaluation_metrics", "security_gates", "tests_are_truth"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2c6245808abdbfd9044781606b291302ed975aca"}
{"id": "train_20638", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "ci_integration", "governance", "documentation"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ea9c267af746c0515192a86ebf69157824a3c260"}
{"id": "train_20639", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "evaluation_metrics", "repo_scale_reasoning", "tooling"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2acaf6eb53d4a01506339fbca83d30228667df53"}
{"id": "train_20640", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "evaluation_metrics", "governance"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "25d7aa2a90d58f0a7b61ee08b80b98d1fecb01f1"}
{"id": "train_20641", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "documentation", "tests_are_truth", "reproducibility"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "05949bdfe468fdbdd8a837e68d5d7e9f0ded2f46"}
{"id": "train_20642", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "tooling", "cost_latency_tradeoffs", "governance"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b4dbc8ba41849c8c9a382db19bf78790360f5a24"}
{"id": "train_20643", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "tests_are_truth", "documentation"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b8159e9145dbd07be4380a9487917fac3de769e5"}
{"id": "train_20644", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "security_gates", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0f21791d2e8ebde8ea3bd2eea086bdd83b614b63"}
{"id": "train_20645", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "ci_integration", "tests_are_truth", "reproducibility"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5c80fc31fcfd42d0a394ba3ae76771e650fec278"}
{"id": "train_20646", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "tooling", "governance"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cbbc925cdabddb07c2e381f69bb4f1165579879f"}
{"id": "train_20647", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "documentation", "governance", "tooling"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bf06b8439c2234b6c9dd4d7167b12050e4e65632"}
{"id": "train_20648", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "governance", "auditability", "documentation"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c78ed3e7b3a03f7b82098c69b6133764c4068f68"}
{"id": "train_20649", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "repo_scale_reasoning", "tooling", "ci_integration"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c70a6f44075c57b3df829065a346f7de60872c24"}
{"id": "train_20650", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "tests_are_truth", "reproducibility", "documentation"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "97d0020dbbf187e0a317782b411d91b263998092"}
{"id": "train_20651", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "governance", "repo_scale_reasoning", "security_gates"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5d2ae2db6a75a67154dc3db0e8ab1c2ac4e99e70"}
{"id": "train_20652", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "evaluation_metrics", "tooling", "documentation"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a06d68a0e65dcbd30b4e248f43121f3496b9f8d4"}
{"id": "train_20653", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "governance", "security_gates", "evaluation_metrics"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "89b8b3ef719f2c33f645e3aeede2ef4d082abf85"}
{"id": "train_20654", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "reproducibility", "governance", "security_gates"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9db463119a1de9cb72611773b1d8ee1a44423af0"}
{"id": "train_20655", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "auditability", "ci_integration", "governance"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7cd81a5b60b59ddc08cc99e2cd9778e8a9ea8442"}
{"id": "train_20656", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "ci_integration", "governance", "tooling"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cdaad25a8fa1abe595ab6cea764f4e1beaf785e3"}
{"id": "train_20657", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "governance", "auditability", "repo_scale_reasoning"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "59aa3643f7d862cacef70dec6d387ba511b3de32"}
{"id": "train_20658", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["auditability", "tooling", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "eb00ef44656a444498b5ebd60334c19daa0d50dc"}
{"id": "train_20659", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "security_gates", "repo_scale_reasoning", "tooling"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e5c73c43104e5ea455079d9285f9ac834ddc5be6"}
{"id": "train_20660", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "repo_scale_reasoning", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "14a5aa0903436eddd5fd7ce1904ecc954721054e"}
{"id": "train_20661", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "tests_are_truth", "governance"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c1fbadff1a3aa3335455e2433319861960c230ae"}
{"id": "train_20662", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "evaluation_metrics", "tooling"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "691b22fabbbaa2aa67389d162ff83e5ed979e805"}
{"id": "train_20663", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "auditability", "evaluation_metrics", "reproducibility"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1a1eb29e4bb3186a9fa0d9d2cea1cbd17199cd6a"}
{"id": "train_20664", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "tests_are_truth", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "51a9e02896e0e15ddf34624529576719d9d32d16"}
{"id": "train_20665", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "ci_integration", "reproducibility"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b01e634298cf5be7a857ad854f7db27c1a266ec5"}
{"id": "train_20666", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "tooling", "tests_are_truth", "security_gates"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "aeae1b9d423b93908de049ca472425121f65bd65"}
{"id": "train_20667", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "tooling", "evaluation_metrics", "auditability"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ca1d850773358e9289832df11f5a0aed3fc1d026"}
{"id": "train_20668", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "auditability", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0970c907721c951b5ad61275587c557069bec019"}
{"id": "train_20669", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "ci_integration", "reproducibility", "governance"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7402c90635fa18f7bb9d1586dc597f3ec5fa11c3"}
{"id": "train_20670", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "tooling", "evaluation_metrics", "ci_integration"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6195cc41ddcb4355b97d5e50b4ba3c350799f4ae"}
{"id": "train_20671", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["governance", "ci_integration", "tests_are_truth", "reproducibility"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f4b3bb34788465f1d8b80a09060c3b78b82d2e34"}
{"id": "train_20672", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "reproducibility", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9f82d37063ca2f40b07479563ff1ddd0725fad8b"}
{"id": "train_20673", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "governance", "tooling", "auditability"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bc47c06cf24af6d938547b59f8b2645aee812b1d"}
{"id": "train_20674", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "documentation", "auditability"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "48e14be1c2135d0145eee3ebacf2b713a337e4e9"}
{"id": "train_20675", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "ci_integration", "auditability"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a3d7aa0abf8560cc6c985d5a606e2fc3678a1baf"}
{"id": "train_20676", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "tests_are_truth", "auditability", "tooling"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "54529cdf0fe3c3b498daaf3bb6b8605d14f9c43a"}
{"id": "train_20677", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "auditability", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0ae24437bb1575743330b5bb5bc5f400db65aa67"}
{"id": "train_20678", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "governance", "tooling", "reproducibility"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c70d358af34231572a0e6af1af622efb0c8e38a6"}
{"id": "train_20679", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "security_gates", "repo_scale_reasoning", "auditability"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e8f0510fd4fc3531e1a1bb757c7b81ca251089b0"}
{"id": "train_20680", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "ci_integration", "tests_are_truth", "documentation"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cd1817a67c588ac4c8a16b78b708b3fce83a27c3"}
{"id": "train_20681", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "auditability", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7aec9963ffee963c226c535df05f76ff7166c9c3"}
{"id": "train_20682", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "reproducibility", "documentation", "auditability"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1583da0da539a0f6b04ca291ce1ef0c390f5ffec"}
{"id": "train_20683", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "ci_integration", "repo_scale_reasoning", "security_gates"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "549d004828fb7c1898b065f8b9079820784fe6d9"}
{"id": "train_20684", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "evaluation_metrics", "auditability", "reproducibility"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a82fd8995d8a78d163fb227ded675001828931cf"}
{"id": "train_20685", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "ci_integration", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5b81edeeece53dbaa70c82537574207c17bd6b6f"}
{"id": "train_20686", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["governance", "reproducibility", "tooling", "evaluation_metrics"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0b62e21a88101113be5d2aaf5123afe0f2268104"}
{"id": "train_20687", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "cost_latency_tradeoffs", "documentation", "evaluation_metrics"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f2dc2ebef4f545f2420338455f4d257999ddae66"}
{"id": "train_20688", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "security_gates", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c14e90d0b3ccc8707ef4f849000460575dec5526"}
{"id": "train_20689", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "security_gates", "tooling", "ci_integration"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b59792a66f35d80d6f7e07bd539b5ae7f1ee9d67"}
{"id": "train_20690", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["governance", "reproducibility", "tests_are_truth", "documentation"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8bba1d55dc69575c4d8bc9e378ae442bca071ce5"}
{"id": "train_20691", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cb201427cbeb6237ef6c2a688c951e5e9e35a54a"}
{"id": "train_20692", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "governance", "auditability"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "01b3dd068b45f8ce322724071e65fcf9dc6ad914"}
{"id": "train_20693", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "cost_latency_tradeoffs", "repo_scale_reasoning", "governance"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "deaaf6267614072475d6907991d5f3f1f080fdd8"}
{"id": "train_20694", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["auditability", "security_gates", "governance", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ec4080310099c11c21c2e3163b2246293362921e"}
{"id": "train_20695", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "tests_are_truth", "repo_scale_reasoning", "documentation"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "00d9a596a453ba5d91ff290d419fed684a2cd82e"}
{"id": "train_20696", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "security_gates", "auditability", "reproducibility"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7c0df2d29dd2390dc0ee845647a4b5d250292c66"}
{"id": "train_20697", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "evaluation_metrics", "tests_are_truth", "reproducibility"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4222a928f1fbf5767c20d5d94ccab828a38add54"}
{"id": "train_20698", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "cost_latency_tradeoffs", "documentation", "ci_integration"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a8f46c7e5f3f6ec9be427b6c6fe1e92893c225c1"}
{"id": "train_20699", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e1b3716a1357a90fbb88f24a811bb16ecd76d07c"}
{"id": "train_20700", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "tests_are_truth", "reproducibility", "auditability"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5b15f92a432768e5b914d71782e442d4459686db"}
{"id": "train_20701", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "governance", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "672e9f4a5f927846dc1669585d9676175bc8aebf"}
{"id": "train_20702", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "repo_scale_reasoning", "auditability", "documentation"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4aa2a38ee97d1b1519e44f2004503d1a588f5e73"}
{"id": "train_20703", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "cost_latency_tradeoffs", "evaluation_metrics", "auditability"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "915640a0bfe795fe4da66a8359f3dcb2888ccbe0"}
{"id": "train_20704", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "documentation", "reproducibility", "security_gates"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "026ca1c8ee3cea588df4dbd1cb8cc25e13743f76"}
{"id": "train_20705", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "evaluation_metrics", "tests_are_truth", "auditability"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9978bb0e05d6a2eca91ae94ba5bd1e24329a1db4"}
{"id": "train_20706", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "reproducibility", "documentation", "auditability"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6accc5c864f1377519675abea3537f7eb645ffd5"}
{"id": "train_20707", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "ci_integration", "tooling", "security_gates"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d66db1a609395cf1dac7441afeb1c8a4cae2344a"}
{"id": "train_20708", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "cost_latency_tradeoffs", "evaluation_metrics", "tooling"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d2fd24050e464465299066212e243c4fd729b0dd"}
{"id": "train_20709", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "governance", "reproducibility", "documentation"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c03ca3603a8078f5daee7f65c6b1a4c065d4057a"}
{"id": "train_20710", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["governance", "reproducibility", "auditability", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "16f5aefbbc2768ad0e36a5d53299d78bae2d8d68"}
{"id": "train_20711", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "evaluation_metrics", "tooling", "ci_integration"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6c5e7ba70394951afd8a1e86d824a520f4d2dc80"}
{"id": "train_20712", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "governance", "tooling", "tests_are_truth"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d557e21fe235028d0e707cdf3ae8c17e1bc7feb5"}
{"id": "train_20713", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["governance", "ci_integration", "tooling", "auditability"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "71332c40d6188c096470ef6b5224c87abf71c6c1"}
{"id": "train_20714", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "ci_integration", "reproducibility", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "65708bebf748ca275fabb103aa632ca480830b5a"}
{"id": "train_20715", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "documentation", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9ed7e24fbf054de36059c3e7fc06ae443fdb70b8"}
{"id": "train_20716", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "tooling", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "64eeeadb6d0b6854cf0e28c351d4ead875d98049"}
{"id": "train_20717", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "76955ffb2f252c7655e2722d95fa72137b1be531"}
{"id": "train_20718", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "evaluation_metrics", "documentation", "reproducibility"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3c95b593fc1d6dcafcdb6740e8c9f95669cda67f"}
{"id": "train_20719", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "governance", "evaluation_metrics", "auditability"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3c6b6324552e978b798e26912a951bcbbcd44552"}
{"id": "train_20720", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "governance", "auditability", "ci_integration"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f56bed9b7ce6e49092da83a6570d63af613bd8b8"}
{"id": "train_20721", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "repo_scale_reasoning", "evaluation_metrics", "documentation"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "617d5e9644b5d5f58e9bd88655b253f0148fe3f3"}
{"id": "train_20722", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "repo_scale_reasoning", "reproducibility", "security_gates"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4b198e67a17425268624cf60538c68886f73113e"}
{"id": "train_20723", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "tooling", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9d8d5d1ee717533064587232cadd72f1ab29cda4"}
{"id": "train_20724", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "repo_scale_reasoning", "tooling", "tests_are_truth"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "223e7781d4a7af911a27617028ea73b1840e1aaa"}
{"id": "train_20725", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "repo_scale_reasoning", "tooling", "ci_integration"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "37ec1f1b6a44e3e47cbbaad5c91d8270dd0d9fde"}
{"id": "train_20726", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "tooling", "security_gates", "evaluation_metrics"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4628b463ca6a385509cc99acd87649f9401b3063"}
{"id": "train_20727", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "security_gates", "ci_integration"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "88261a85b059501f759883f82c7444973ac258fd"}
{"id": "train_20728", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "tooling", "tests_are_truth", "reproducibility"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "856cd150b84836264469c2704defe8ba172bc2f0"}
{"id": "train_20729", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "cost_latency_tradeoffs", "security_gates", "auditability"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d1371523d0e3607b7a873aa3b3ca60c3187ea7e2"}
{"id": "train_20730", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "security_gates", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "15c103ba376755d289d36fc591c533f4dcc1bc49"}
{"id": "train_20731", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["auditability", "governance", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ae556a09e86d9170118acc22dd349363428fc783"}
{"id": "train_20732", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "security_gates", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3e9c29839dfd37aa3678d4a67ab5116f4a53a2e9"}
{"id": "train_20733", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "documentation", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c0593b036da998ceb3cdf6f38229f3550b6522db"}
{"id": "train_20734", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "documentation", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "489b4210a05cf21177fb343ba05b038c1aabec7c"}
{"id": "train_20735", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "ci_integration", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "221d802af38eb1cd86fce9f34f66275c22c32a80"}
{"id": "train_20736", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "reproducibility", "repo_scale_reasoning", "security_gates"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "721731e2b9160beb4f329626808cd4fc88afccd5"}
{"id": "train_20737", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["auditability", "governance", "security_gates", "reproducibility"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7f38d82bbfc56caac17f73a7d775796a1fc6e9c1"}
{"id": "train_20738", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "tooling", "tests_are_truth", "auditability"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "78111ccd211f16c6002b1547cade30c5a97ce9dd"}
{"id": "train_20739", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "security_gates", "documentation", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9b9a014361e4607907d9fcb60419a031d7ae6dd5"}
{"id": "train_20740", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "governance", "tooling", "security_gates"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9c86a6c2e19bbc86e9711fb63593394cf29eb27c"}
{"id": "train_20741", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["auditability", "security_gates", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ff59803c2eca85123f794a75cdfdfde3d586dcee"}
{"id": "train_20742", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "evaluation_metrics", "ci_integration", "reproducibility"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "190303f9d6dd3327407fe1fcd8cd9d2b20a66998"}
{"id": "train_20743", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "evaluation_metrics", "governance", "documentation"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0d804f9abe36074cda88a1465141af707496f524"}
{"id": "train_20744", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["governance", "tooling", "reproducibility", "ci_integration"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c1efd76f690c7ef7de67e9b20387973f3696ec9d"}
{"id": "train_20745", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "auditability", "governance", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5265a0bc44fc1b12b8fcb8d2063f30518bbdd83a"}
{"id": "train_20746", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "auditability", "reproducibility", "governance"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "630e0cf864bfb93dbc3491dedb0319742a6fba8a"}
{"id": "train_20747", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "documentation", "auditability", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a62ef85c5d859b0083d1b12dceaa09a916960d3f"}
{"id": "train_20748", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "tests_are_truth", "repo_scale_reasoning", "tooling"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d11301a105adc727e13935aa8ff11a856c0b63f5"}
{"id": "train_20749", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "governance", "security_gates", "documentation"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4d1ea854fe6e39005e7c249987a163a5aa8e436e"}
{"id": "train_20750", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "ci_integration", "security_gates", "tooling"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bba8c1b2375baa79b25d8b68a8416c88ae0a1a5d"}
{"id": "train_20751", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "cost_latency_tradeoffs", "security_gates", "governance"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "90175149607c89c0c20c0088e9c22577d6573af6"}
{"id": "train_20752", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "security_gates", "ci_integration", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d9ce05fda1f0bd8d87b1ad59016d583307124232"}
{"id": "train_20753", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "governance", "evaluation_metrics", "tooling"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b9088aafa87b7814a754d27d020f7254e42d0864"}
{"id": "train_20754", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "tests_are_truth", "tooling", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6700303413640188721295bc1899d12ee5aeb954"}
{"id": "train_20755", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "tests_are_truth", "ci_integration", "security_gates"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "27574483c0e3f5f3c221a07971897e65bd207d47"}
{"id": "train_20756", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "auditability", "evaluation_metrics", "governance"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "979181b56c409a88dfc9cbce96fd4a9c34d3a6c8"}
{"id": "train_20757", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "cost_latency_tradeoffs", "tests_are_truth", "reproducibility"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0fc249e1da72bec3cd6dde33f6856b19e0d9ba17"}
{"id": "train_20758", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "tooling", "security_gates", "governance"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e3076447d6116dbae7efa569ee71e262631f2ebc"}
{"id": "train_20759", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "ci_integration", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bb64f43f805e0d7640f4ffe9a0c5338ee60bd8fd"}
{"id": "train_20760", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "cost_latency_tradeoffs", "evaluation_metrics", "governance"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7afdf89492526efbbcbefe940760b5d9b54c36e8"}
{"id": "train_20761", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["auditability", "tooling", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ec023554f72bc7a11533baaec74f01f76c721be7"}
{"id": "train_20762", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "ci_integration", "tests_are_truth", "reproducibility"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c41419c5eba8fa123e46e4e6524c739b7fc6a555"}
{"id": "train_20763", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["governance", "repo_scale_reasoning", "reproducibility", "tests_are_truth"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8e9fc70edb7b7ad72701071a30d796221dc9d443"}
{"id": "train_20764", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "security_gates", "governance", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9650bf9e318968463320a78eac5afe00dfcb6b60"}
{"id": "train_20765", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "tooling", "security_gates", "auditability"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b7de1d117a9c9b55e37bb36360572d5d0fee614a"}
{"id": "train_20766", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "evaluation_metrics", "ci_integration", "documentation"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f0b94d00628cf9e82d665ed961e9e32a8f33e85a"}
{"id": "train_20767", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "security_gates", "reproducibility", "tooling"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7e2341e2df02f7143282db0562a37bdd2f2f5bde"}
{"id": "train_20768", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "ci_integration", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6ff2dad90e1e1276d3ddc93b22a603c11fc33cf8"}
{"id": "train_20769", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "ci_integration", "security_gates"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b3d02dfc330a739aa56e7d438c43ca3ee9113e81"}
{"id": "train_20770", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "tooling", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "496d95f76d8a4412af0d1f426e8a7b3d08c7e722"}
{"id": "train_20771", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "ci_integration", "tooling", "governance"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "53288cc75a99698bb12cbc0c336bd67c3a896886"}
{"id": "train_20772", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "ci_integration", "tooling", "governance"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "324b40e4020a412b2da0d8731b0b107977d302fb"}
{"id": "train_20773", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "governance", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "56d68f00b629e6d8087b9968114f190060e2e437"}
{"id": "train_20774", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["documentation", "evaluation_metrics", "security_gates", "tooling"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6a819d7d2345aea3d5e6d8c08516baf6095763f5"}
{"id": "train_20775", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "ci_integration", "documentation", "reproducibility"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6757200439f8f444c86e534cb64b75ff62e0c358"}
{"id": "train_20776", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "21710c2e5e2bfa14ac76ab06ac8ccd880539c3d6"}
{"id": "train_20777", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "cost_latency_tradeoffs", "governance", "tooling"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9afa47e8f9962338525d51555d2e45659a124bb8"}
{"id": "train_20778", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "auditability", "documentation", "security_gates"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "10ed97d645168d440dfbd9e2a95a6808df94dc87"}
{"id": "train_20779", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "evaluation_metrics", "security_gates"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3f610c852a4140d0237cda85534561b0f7ab8e03"}
{"id": "train_20780", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "documentation", "security_gates", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "982f41159f39bf4fd113f8ff143a02353a639893"}
{"id": "train_20781", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "auditability", "documentation"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "46628ac18c7bbadf99b31edbac3ce6781a96d4a2"}
{"id": "train_20782", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "repo_scale_reasoning", "auditability", "reproducibility"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dfde98547c62e897353db2ce00e037edadab8238"}
{"id": "train_20783", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "tooling", "governance", "documentation"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c4dafbb0b8c5adaa25182f25e2ddd621a88dd80e"}
{"id": "train_20784", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "tests_are_truth", "reproducibility", "tooling"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ef257f193ef93ba6817239ad9d878c20be489346"}
{"id": "train_20785", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "auditability", "documentation"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "178bd59d7b1bd4780ef6bfd094279dd8e43833a4"}
{"id": "train_20786", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "ci_integration", "governance", "tooling"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4708e0b2bc32a49c3c79170cb4078cc629a9b786"}
{"id": "train_20787", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "auditability", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7e5d55699228e279ea75494353e9daba51a66a50"}
{"id": "train_20788", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "reproducibility", "ci_integration", "documentation"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bea4c214cf1a535cf8e3bfb63d9638eeb8a485dd"}
{"id": "train_20789", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "tooling", "security_gates"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5059137b0be752a253b0f767e94ae18c8676ae37"}
{"id": "train_20790", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "auditability", "ci_integration"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "03104706cf80a4e768bdd597814935dcb5e53654"}
{"id": "train_20791", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "governance", "reproducibility", "documentation"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1acf97d6d884c95d20272f4a27c0bb6ae260c162"}
{"id": "train_20792", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "security_gates", "tooling", "ci_integration"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5f740b259d0cbf946d86b10f1bd805c985336926"}
{"id": "train_20793", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "tests_are_truth", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d078c570307305a501746be2d5f05463b43d2aaa"}
{"id": "train_20794", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "documentation", "reproducibility"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bc7a8454d4bfe4080131ac1a52ea78e9c0bf3f4a"}
{"id": "train_20795", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "security_gates", "ci_integration", "tooling"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4235c0a47708143c2dcc01d24e29796e1772caa6"}
{"id": "train_20796", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "documentation", "repo_scale_reasoning", "governance"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "81348d8259fcea6fb92100f39f2eaed115d36073"}
{"id": "train_20797", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "tooling", "ci_integration", "security_gates"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b50574c26689ca0e02bc26cf2b5f94af3cdd3c60"}
{"id": "train_20798", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "ci_integration", "tooling", "documentation"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2c95a113abfa8ea15d2f6365f17334520e684201"}
{"id": "train_20799", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "repo_scale_reasoning", "auditability", "reproducibility"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4444e8356dbb20c48683458e92d3143465c889fd"}
{"id": "train_20800", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "auditability", "reproducibility"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "823e35fa0431c38a2c8d4b867b528edca25490f7"}
{"id": "train_20801", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "security_gates", "evaluation_metrics"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7264bfeedf3bef8711ebdb33f57363090d7184e7"}
{"id": "train_20802", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "repo_scale_reasoning", "governance", "reproducibility"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "21247bf6ec5b436158de9047059e8f59fb2c5d80"}
{"id": "train_20803", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "security_gates", "governance"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "32820d4cbb49aeb6bd537cd4f324995c58ccc88c"}
{"id": "train_20804", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "cost_latency_tradeoffs", "ci_integration", "governance"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3b1ea30b5874bb5b6e68d1c2117f3325451d9e3d"}
{"id": "train_20805", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "repo_scale_reasoning", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b6b2927c385d5258018c9605374c072a4f68ffa3"}
{"id": "train_20806", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "auditability", "evaluation_metrics", "documentation"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2e96c292e8e4870b3f1196c085c8d06b0ddc2535"}
{"id": "train_20807", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "documentation", "ci_integration", "tests_are_truth"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9e849179b75021ed34cb0e8d765264e1f173f0ca"}
{"id": "train_20808", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "tooling", "governance", "security_gates"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "660e263b4a0fd14dd2aa39e1da2e0a75aca903ba"}
{"id": "train_20809", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "ci_integration", "security_gates", "auditability"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "efdf2ef471ac0d9c352c38fa5ace2635a12e32f2"}
{"id": "train_20810", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["governance", "auditability", "ci_integration", "reproducibility"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fae0056bd2b726f001aa7bc94b2ac72c5614e3cc"}
{"id": "train_20811", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "ci_integration", "reproducibility", "documentation"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0c5d653801ef47914cdec172708e762438180ba9"}
{"id": "train_20812", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "documentation", "auditability", "evaluation_metrics"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ca04357f9ad03ca6d8b7657b7a35c4efd49e6f17"}
{"id": "train_20813", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "security_gates", "documentation", "reproducibility"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "022dc1f80255788a950a7dc202c457914449cb64"}
{"id": "train_20814", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "reproducibility", "cost_latency_tradeoffs", "security_gates"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "96eaba6c27ecd6469706a7b024a803a7ceff369e"}
{"id": "train_20815", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "evaluation_metrics", "repo_scale_reasoning", "auditability"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ce94ab02585ff8ff35b315cf843dec6805259da6"}
{"id": "train_20816", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "tests_are_truth", "documentation", "security_gates"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "954fb11364ad4edf2da543875fb4876846bdd6c8"}
{"id": "train_20817", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "documentation", "repo_scale_reasoning", "security_gates"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d40763ffec00229e0977f661ccff8f9cbd0b35be"}
{"id": "train_20818", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "ci_integration", "security_gates"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "53e1d01d7d9396b05698f26035c6ab3c9584772d"}
{"id": "train_20819", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["governance", "reproducibility", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "331b209a41b72f07d40a6bf5bb5abd943ae59b1d"}
{"id": "train_20820", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "tooling", "documentation", "auditability"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bc8d3a58167c91ac55e56e607c3fca53f269ca48"}
{"id": "train_20821", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "auditability", "tooling", "tests_are_truth"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4a5121faa79310d7b5a58cc758560e493f547227"}
{"id": "train_20822", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "security_gates", "tests_are_truth"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2bdf5f454e32fb6f10b4b58e437135d728821527"}
{"id": "train_20823", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["governance", "repo_scale_reasoning", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cd89096d2f0ea12f2962c862c8027347606d8f5a"}
{"id": "train_20824", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "tooling", "tests_are_truth"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "edac475c79f36bf651035836c7712d9ea1ec4ad9"}
{"id": "train_20825", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "ci_integration", "cost_latency_tradeoffs", "governance"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "474a061b8e10abdc5be295786429e0c273eed04e"}
{"id": "train_20826", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "tests_are_truth", "governance"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bde13b44e38bdcd4e84b3e87435ada12e04ace0e"}
{"id": "train_20827", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "tooling", "reproducibility", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ac1f9abdf028f35042561f4c553d8c657d480bd3"}
{"id": "train_20828", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "tests_are_truth", "ci_integration", "documentation"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1710d3377eafe912bdefe67ca195512f55bdbb61"}
{"id": "train_20829", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "ci_integration", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7b825d1fa57259987c0a326d9df16747206bbb66"}
{"id": "train_20830", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "reproducibility", "ci_integration", "governance"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7f86a160aa999d9aa5554e0cddce18f1bab9c902"}
{"id": "train_20831", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "documentation", "tests_are_truth", "security_gates"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "784e2de645a418e477a38ba2c4aa7294caa8c09b"}
{"id": "train_20832", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fc236fb033b32d1c6fbc7be6a297e95a33f2089f"}
{"id": "train_20833", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "governance", "security_gates", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a889278fe733b9694cc40d6b0cd7944f85edfb61"}
{"id": "train_20834", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "documentation", "tests_are_truth", "governance"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7ed6e37c5a05946d7f0ec775e7d1e47746b49d40"}
{"id": "train_20835", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "auditability", "tooling", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "14b2e488b0c2388d052a9ec2ff9c531a1a774997"}
{"id": "train_20836", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "security_gates", "governance", "evaluation_metrics"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bd0677105d3c4793f4a64af142eb5a3558caabab"}
{"id": "train_20837", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "security_gates", "auditability"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5f8991f6510d681425ac31277cef23a37e1d6a94"}
{"id": "train_20838", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "documentation", "tooling", "security_gates"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3d67f09967553c4e83e810343ccd8f161db5eabd"}
{"id": "train_20839", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "tooling", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0539173409c183fdf9e16c7fffd41f2eac133fd1"}
{"id": "train_20840", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "tooling", "reproducibility", "auditability"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "70b11b84be0213249f39ff2d20b2e07372c4a91f"}
{"id": "train_20841", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "documentation", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fa72155f431139b68e2fa4d0408d21884ce85117"}
{"id": "train_20842", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["auditability", "documentation", "governance", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f81936f3b32b1f696e214084ecfe7ec9aa8697c1"}
{"id": "train_20843", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "auditability", "repo_scale_reasoning", "security_gates"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "70e10bc5e020eef0a4d733495ed27d5c80f36d2d"}
{"id": "train_20844", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "ci_integration", "evaluation_metrics", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bdb7fe1e39859a31182875d0419887c111db9de2"}
{"id": "train_20845", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "reproducibility", "tests_are_truth", "documentation"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e19fb8a484417dc22a6f20d22bc2c02023363eb8"}
{"id": "train_20846", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "repo_scale_reasoning", "documentation", "tests_are_truth"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f7f7e592c0923a93cbf11dc2bd8c0ddb35977988"}
{"id": "train_20847", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "documentation", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7fcb9771a9d0ec056b5f0b723d5818e4d391123a"}
{"id": "train_20848", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "governance", "ci_integration", "tooling"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2123748d05d8c085b27984267e0a3b3c17dec203"}
{"id": "train_20849", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "governance", "security_gates", "documentation"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "58fd9e4b99dbb84dd3226443cf549e2263a1c4a7"}
{"id": "train_20850", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "evaluation_metrics", "ci_integration"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d4a22f19a6be7d35e9646b8a8e4f956c9af0ae72"}
{"id": "train_20851", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "tooling", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cf0f1be476f9ca6bc935715790f8c41fdbff8950"}
{"id": "train_20852", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "evaluation_metrics", "governance", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5ee019972aca470f3a5c7b7fb22f70a93e7112a3"}
{"id": "train_20853", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "cost_latency_tradeoffs", "documentation", "security_gates"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0746edde06a4dec378e1fef6f9bcccaaeaaeb10c"}
{"id": "train_20854", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "tooling", "ci_integration", "tests_are_truth"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "872d8a6a3725a46a00a60006560622f771234ab6"}
{"id": "train_20855", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "auditability", "governance", "reproducibility"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ed39c76127142ffa6ac2b2f63b534cf4d1f7e6ca"}
{"id": "train_20856", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "reproducibility", "security_gates", "documentation"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "01d5546f1f2cdba363902dbee0016dfb2978c6bf"}
{"id": "train_20857", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "ci_integration", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "551e3bd57fa5d809adecd38f1505de7877bfd486"}
{"id": "train_20858", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["documentation", "ci_integration", "auditability", "governance"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b9db2a6d67881bd38698c69a9a3353b0c7835361"}
{"id": "train_20859", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "reproducibility", "evaluation_metrics", "governance"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a66f708957a4d41625dfe0c16f0f627d1227c2e6"}
{"id": "train_20860", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "tooling", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bfe54c1f9d3e8160c8a35bcbcdee2223230f4bb3"}
{"id": "train_20861", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "auditability", "evaluation_metrics", "tests_are_truth"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0fd8a3a3c11c6cb07ebfe8d83a2ad1ac18214e33"}
{"id": "train_20862", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "evaluation_metrics", "tooling", "auditability"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "67970d5a866fdefbf7c54573204c413fae65a998"}
{"id": "train_20863", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["documentation", "reproducibility", "repo_scale_reasoning", "auditability"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5032ef50ba56b1f10458f3cfad34f49964ce708d"}
{"id": "train_20864", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "auditability", "governance", "ci_integration"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5304a7eeefc19c1f379a537780fb749de929bc91"}
{"id": "train_20865", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "reproducibility", "ci_integration", "security_gates"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8bf413822fc67489bab6b02b80f8a836b03c464f"}
{"id": "train_20866", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "ci_integration", "governance", "tooling"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1624f7c88da44cfedd6fa1a6f64df2566dd67777"}
{"id": "train_20867", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "evaluation_metrics", "auditability"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8d4cdb322d76eb3bf8d65cceda4b87c6ebdf9b5a"}
{"id": "train_20868", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "documentation", "governance", "tests_are_truth"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2948753abc9dba3ccbf218cb5d27f8d2dde98d05"}
{"id": "train_20869", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "tooling", "reproducibility", "ci_integration"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "afc9946ba38105fbf9d571ab24ef8831bd28ac05"}
{"id": "train_20870", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "documentation", "security_gates", "reproducibility"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "98f5c0c78023d2d3e9c1ecdd6028875e30e9587a"}
{"id": "train_20871", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["governance", "auditability", "ci_integration", "repo_scale_reasoning"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "efdbe375940aad0b7f7d86a7e45b5d8a80a22270"}
{"id": "train_20872", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "repo_scale_reasoning", "tooling", "governance"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a1b7288e871d2c69fb2ba1f3ff3fb45478bb2d5f"}
{"id": "train_20873", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "documentation", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2477e8733776ce0c4d8aeec9a8e82c896eb1dbd6"}
{"id": "train_20874", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "repo_scale_reasoning", "evaluation_metrics", "tooling"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a5c1a13dc3cd6cb61d56b37410292057d1da130b"}
{"id": "train_20875", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "security_gates", "tests_are_truth"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f7603927e5a2ffe1c7fa3c65abc2b76f04acfc50"}
{"id": "train_20876", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "ci_integration", "tooling", "reproducibility"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "47bd91644d9e19bb33037a3fa9da40e8a2748502"}
{"id": "train_20877", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "ci_integration", "documentation", "auditability"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "20591350ba22e218324ac022077b78d4bf3efe87"}
{"id": "train_20878", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "ci_integration", "reproducibility"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "62e38c5ddf0d6efbb57f517df645f43c22d9cc7e"}
{"id": "train_20879", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "reproducibility", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "551a5598f8e034c7abbb79aa898b6dab7b6179b6"}
{"id": "train_20880", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "documentation", "auditability"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6e055efe3a1fe4a617e27f8462e595dd85ba35fa"}
{"id": "train_20881", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "evaluation_metrics", "cost_latency_tradeoffs", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "25fc3166ef5e14907d2e7b1caa8bc500c8c0e2f9"}
{"id": "train_20882", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "governance", "evaluation_metrics", "tooling"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8df91c2c297fdebb6aeefc5a652edacc9c19a030"}
{"id": "train_20883", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "tests_are_truth", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4364f9cec4413ac45737209e3781b07b2a4a026c"}
{"id": "train_20884", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "governance", "auditability", "evaluation_metrics"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5cff20470d7a6524be4539d9f1f1f99e3ab9de56"}
{"id": "train_20885", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "tooling", "documentation", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9cc8e6e4ddf7969328e8080e305078b7f14ffedb"}
{"id": "train_20886", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "evaluation_metrics", "governance", "tests_are_truth"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a5fdbed797bc051134daaf3254bcef9e137a2731"}
{"id": "train_20887", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "evaluation_metrics", "security_gates", "governance"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "68078c3e934139d4e9e0635a36a803357118c2d0"}
{"id": "train_20888", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "auditability", "tests_are_truth", "security_gates"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "943e7e7a877ec7530f02d421d1b6dea98222dd96"}
{"id": "train_20889", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "repo_scale_reasoning", "tests_are_truth", "auditability"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a10d38d1b6973eeee21954305c5b0d0e58e6d543"}
{"id": "train_20890", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "governance", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "78af0edfb43036c00b7c428a9b464083a1035243"}
{"id": "train_20891", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "evaluation_metrics", "ci_integration", "governance"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a3c7795df521d0258eb0d4d0f36367d73258dae0"}
{"id": "train_20892", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "ci_integration", "governance", "tooling"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c828f6fb87d0a00a1d65e31543cf1c826ef7da98"}
{"id": "train_20893", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "documentation", "tooling", "governance"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "18e03ecd3a304fb80195b22b67b8b62270bfb9e4"}
{"id": "train_20894", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "repo_scale_reasoning", "tooling", "governance"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2ce4e056479d7724cc517ac11ead52546114a7af"}
{"id": "train_20895", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "tests_are_truth", "governance", "repo_scale_reasoning"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bf64c120fef56e278e0e816703be07b4e9f54d6b"}
{"id": "train_20896", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "repo_scale_reasoning", "security_gates", "auditability"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "58fd9e4b99dbb84dd3226443cf549e2263a1c4a7"}
{"id": "train_20897", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a852e8c403e2a0d9ee721ad6c6714e1a40af01e0"}
{"id": "train_20898", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "auditability", "cost_latency_tradeoffs", "security_gates"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "16dd92e4930822565877b171b72da9567a310e2b"}
{"id": "train_20899", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "cost_latency_tradeoffs", "tooling", "auditability"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c80147d88cd04674969dc7bb8fe53b5f262a36c1"}
{"id": "train_20900", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "repo_scale_reasoning", "evaluation_metrics", "auditability"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "455e34e0adfb8e116bdceb4e1bfac8b7c490a801"}
{"id": "train_20901", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "governance", "documentation", "ci_integration"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6d51b6c72869e9fafe4fc80762b8e1c079e5111f"}
{"id": "train_20902", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "governance", "security_gates", "auditability"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "294480eb6c6a898ee9bd78c400e46615e6ae58d0"}
{"id": "train_20903", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "documentation", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f7a138b47adfd644105bbaf5d22b249ed643859d"}
{"id": "train_20904", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "tooling", "cost_latency_tradeoffs", "governance"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "296427a06c2bb7a98432afc9bfb73bc75ea083f3"}
{"id": "train_20905", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "reproducibility", "security_gates"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c3499540067e468d26c812383a2832479b7150f2"}
{"id": "train_20906", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "repo_scale_reasoning", "security_gates", "documentation"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d8cf4fcf66b7e460ac85962b558aa2ea5d612ecf"}
{"id": "train_20907", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "security_gates", "reproducibility", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "568f52ccb5e4f4a08e3223f3e56c2fad0de22520"}
{"id": "train_20908", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "evaluation_metrics", "tests_are_truth", "governance"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "15aba323b433d3892e11cf70a5a7f2fbe677cb5d"}
{"id": "train_20909", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "cost_latency_tradeoffs", "ci_integration", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b1274ef6a56b6be5dcc2b83c1a463a58364b26f2"}
{"id": "train_20910", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "tooling", "ci_integration", "governance"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0313224fe3bdffc5bf6a6ab5ef0e384b7d52df52"}
{"id": "train_20911", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "auditability", "evaluation_metrics", "documentation"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b7a71777b7c79019b76fbd2c36a3394af47e68d1"}
{"id": "train_20912", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "evaluation_metrics", "tests_are_truth", "ci_integration"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ecb0386a093e6a438f3218a9a824862bf15a8d96"}
{"id": "train_20913", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "governance", "security_gates", "auditability"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "43cde36c5acdc033efec5be60f65e5a564d967a1"}
{"id": "train_20914", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "ci_integration", "governance", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7ef1350fcc49f6d59c1e429da482f5f669dc6b13"}
{"id": "train_20915", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "ci_integration", "reproducibility", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dd0b3e5c21108e4dbf09c30830b468f85cd6ea47"}
{"id": "train_20916", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "repo_scale_reasoning", "reproducibility", "security_gates"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d32de29cc2ad659114866302c47e19b05f3eb927"}
{"id": "train_20917", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "governance", "repo_scale_reasoning", "auditability"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0a859652b0e69c0cc65ad848a841ccdffffe25c3"}
{"id": "train_20918", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "reproducibility", "auditability", "documentation"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5d3c8b50b80176bb3c1b17922401e3c08f48f40b"}
{"id": "train_20919", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "evaluation_metrics", "documentation", "tests_are_truth"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6f10730ac326434c64e73d2b75d65f51cfd8ff3f"}
{"id": "train_20920", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "repo_scale_reasoning", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e2436c48dfbde4d84ee2f1048ed52f2d9279016b"}
{"id": "train_20921", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "documentation", "cost_latency_tradeoffs", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f193ba1c4ba36193cc932410cf95751ebda7baad"}
{"id": "train_20922", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "security_gates", "evaluation_metrics", "tests_are_truth"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "53985bcf3895d022c5d4aca4b27648c447aca7b7"}
{"id": "train_20923", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "evaluation_metrics", "reproducibility", "tooling"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ed2d50b371cadb72948c1f4045acd669cac668a7"}
{"id": "train_20924", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "documentation", "security_gates", "tooling"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "45eb02cd139f3f032e37df466f2a157044bd9e98"}
{"id": "train_20925", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "documentation", "evaluation_metrics", "ci_integration"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e336cb787159ed21c72c7cf9e52e1ccacecba5ad"}
{"id": "train_20926", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "security_gates", "governance", "tests_are_truth"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9584ef0b3da2e07c2a73f8eec851066353a57acb"}
{"id": "train_20927", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "repo_scale_reasoning", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "354dd96595821e1a95ef0afb71dcaeebf00a6fec"}
{"id": "train_20928", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "governance", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "18c505533fa6ccba312505f65a7a7e275e447956"}
{"id": "train_20929", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "auditability", "repo_scale_reasoning", "reproducibility"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "112844fbbf4ac1742c1208cd677a11dc72839ae4"}
{"id": "train_20930", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "security_gates", "governance", "tests_are_truth"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5f394154f0bc6a0000798f6bafaf21cc934e160d"}
{"id": "train_20931", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "tooling", "security_gates"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c1e615db5203d3d9edc8081ad370789d73791f1c"}
{"id": "train_20932", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "tests_are_truth", "auditability"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c88eb69f0f77bdcd1a2e4c5482dc18bb3fc492b8"}
{"id": "train_20933", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "governance", "security_gates"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ba11474eecf7d916e1d815a42576e6802531282f"}
{"id": "train_20934", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "documentation", "auditability", "governance"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "39aa1bea7a45d8d150801e87b6b38426c4e2e0e5"}
{"id": "train_20935", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "auditability", "evaluation_metrics", "tests_are_truth"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ad49338b7f8ef8f42b778c64f44f3cf9ae62296d"}
{"id": "train_20936", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "tests_are_truth", "tooling"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f48cb39e0dba2e45b8216fce58250cf7a5437406"}
{"id": "train_20937", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "auditability", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c2e198dfeabb71be9ac47bd7d6750afde8738511"}
{"id": "train_20938", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "governance", "auditability", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b7ace81c6496f187cc7b17a9ff5f489d19bae2a9"}
{"id": "train_20939", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "documentation", "governance", "tooling"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "27104f9f9ba002dc3348abdaeadd636b02e385c8"}
{"id": "train_20940", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "reproducibility", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a805c85028bec6667371ac8864f597831a9043fe"}
{"id": "train_20941", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "auditability", "governance", "ci_integration"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d5bb0543808c42546840d17029c3de19918203ed"}
{"id": "train_20942", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["governance", "reproducibility", "ci_integration", "auditability"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ebdf2da374b96f062706d6b58f048b0468ec5a95"}
{"id": "train_20943", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "documentation", "governance"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9ccb16b82f5f956ac4df0420d191f65ef740ab94"}
{"id": "train_20944", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "security_gates", "tests_are_truth", "governance"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c11444599bb8a86cc409a791fe8aba950a94adf8"}
{"id": "train_20945", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "reproducibility", "auditability", "governance"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e18d4934de62a2ed8a42948597f120493366d6f5"}
{"id": "train_20946", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "reproducibility", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e684dbc9b77cf0fafe6b8d1ba37bbdd5ce44a140"}
{"id": "train_20947", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "tooling", "security_gates", "ci_integration"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "929bd47d24d00821a12858b4fa74103657f4b922"}
{"id": "train_20948", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "ci_integration", "evaluation_metrics", "governance"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "aee61fba226a8562d98e0a2592eafa8383b69247"}
{"id": "train_20949", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "cost_latency_tradeoffs", "tests_are_truth", "ci_integration"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9f7a9dfb7b73a71e4489fa3f5bb84078efe3a896"}
{"id": "train_20950", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "ci_integration", "tests_are_truth", "tooling"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d3a151c6667c98413d7f67f45b2d0088391e9e45"}
{"id": "train_20951", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "governance", "tooling", "evaluation_metrics"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "466e9757e26f0978329dd48859985eeeceb9ff49"}
{"id": "train_20952", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "tests_are_truth", "evaluation_metrics", "ci_integration"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f5940c61c9b9bf4d92ac13c7db9e0386a1777ec0"}
{"id": "train_20953", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "security_gates", "documentation"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a20987978395ee6fc98e1b16e5e725d93aac3f28"}
{"id": "train_20954", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "repo_scale_reasoning", "tooling", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3a85be7f2d062d6e9ad2e8bdae3ec26dd338ee37"}
{"id": "train_20955", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "tests_are_truth", "evaluation_metrics", "auditability"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2dc4d92e95c2b3c05ab2738afd337683d9580226"}
{"id": "train_20956", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "tooling", "ci_integration"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9a141d2051179ad6d309055adad3ed2ce58dae5a"}
{"id": "train_20957", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "tooling", "documentation", "reproducibility"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ba59e24cc6cbffc08d653b91303b94866cbf3fcc"}
{"id": "train_20958", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "reproducibility", "ci_integration", "documentation"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2000560a3532c59d9e88e3fd110df792183f5237"}
{"id": "train_20959", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "tooling", "auditability"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "53510ee5ab4ab8d1971cf0e0839b7aedf9a6ff49"}
{"id": "train_20960", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "tests_are_truth", "ci_integration"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b84a515a0ab83576db43463451ae835805af2478"}
{"id": "train_20961", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "ci_integration", "documentation", "auditability"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9b64891b0482db718510eec8a8c956978cc297f7"}
{"id": "train_20962", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "ci_integration", "reproducibility", "evaluation_metrics"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8dae2cbf3199131d1d5fa3de39d8815d16beeeaf"}
{"id": "train_20963", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "tooling", "documentation", "auditability"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2e494702397a5b3200556ad7abfd14d439dcb960"}
{"id": "train_20964", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "auditability", "ci_integration", "documentation"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "29bac71755cf8b488e33ca97eae27e35f804878c"}
{"id": "train_20965", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "tests_are_truth", "tooling", "evaluation_metrics"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f4370ad6f8c16ecb2404e3219610a468c4698ed4"}
{"id": "train_20966", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "documentation", "auditability", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2dc710b53c9497a54ea9bbb985c940187472211a"}
{"id": "train_20967", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "auditability", "documentation", "repo_scale_reasoning"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "eaa0647f35ad3de265709a764f035dca23126211"}
{"id": "train_20968", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "security_gates", "evaluation_metrics", "governance"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "80795618c8466388ba804935faa05ef3b298f84a"}
{"id": "train_20969", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "auditability", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "23fb4ff56c0a3a26d106ed230ecdb3feb284ce3a"}
{"id": "train_20970", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "reproducibility", "tooling", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ab5e984b5bf152d06fc91ad8c944fc67cf5ba69c"}
{"id": "train_20971", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "repo_scale_reasoning", "tests_are_truth", "security_gates"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2d3587d3deed4a1eea26a8b68521be2df9554419"}
{"id": "train_20972", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "ci_integration", "governance", "auditability"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f9eae8f8e534d22d5ce259815d8b60007f4eef72"}
{"id": "train_20973", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "reproducibility", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0fbfe7603e86439b1c738f79917948ac85100460"}
{"id": "train_20974", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "tests_are_truth", "auditability", "documentation"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6c8e9b6f07265c754e6050c502c55e7249c81481"}
{"id": "train_20975", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "auditability", "tooling", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b3759674716b4bef1c5d64e7e427b521852bcd38"}
{"id": "train_20976", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "auditability", "repo_scale_reasoning", "documentation"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b712f9de0be6c4134c9e9fe39ca31720b5ff53c9"}
{"id": "train_20977", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "cost_latency_tradeoffs", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "99d6cc95d74008f61c697773d691ee87e75732f4"}
{"id": "train_20978", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "auditability", "evaluation_metrics", "reproducibility"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "eb66fcdb4eabf1b1273437c15baccbd257e836ef"}
{"id": "train_20979", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "governance", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "72fa2b05cea9907365e309af4c3b3acb674b6101"}
{"id": "train_20980", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "ci_integration", "repo_scale_reasoning", "governance"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1a66b8bb352d728e75eedf70f2183d06e3ee9ab2"}
{"id": "train_20981", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "ci_integration", "evaluation_metrics", "governance"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "956e8b4b27e4b2f148ff1aaba14d1ce204b05865"}
{"id": "train_20982", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "auditability", "ci_integration", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fbb2f13a0aa8b2e5303e0b3a8f2133d7a91178e5"}
{"id": "train_20983", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "evaluation_metrics", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cc76bfd73d0798278e1b76b4f593706f6fc64103"}
{"id": "train_20984", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "tooling", "auditability", "evaluation_metrics"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "96612f77aea42fb7a6e5709cbfc775349d9355b9"}
{"id": "train_20985", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "repo_scale_reasoning", "security_gates", "documentation"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "70f4af87dfde7b9a32db19f0b8e07d7fdf98bc46"}
{"id": "train_20986", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "auditability", "tooling", "evaluation_metrics"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7d15a8a13d8d248a10486bcdb49d11940cf20b0f"}
{"id": "train_20987", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "auditability", "reproducibility", "ci_integration"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c46bf7619015e4ffa107dcf8c65bf84568396743"}
{"id": "train_20988", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "auditability", "ci_integration", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "01eaea649004847db8de02db65029fdb3a2945bd"}
{"id": "train_20989", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "auditability", "governance", "documentation"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "de44aca55644e999dd3d4758c6a0b3d12117a995"}
{"id": "train_20990", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "tooling", "tests_are_truth", "auditability"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "544e455dceac3e1f075717b2401f2870fd857f69"}
{"id": "train_20991", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "auditability", "tooling", "governance"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "47163cc18a17d7ae41ae866cee9e469c7f7c07a4"}
{"id": "train_20992", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "repo_scale_reasoning", "documentation", "tooling"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7a9783cea5d90177a5b791d64127bdf238d3ff34"}
{"id": "train_20993", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "documentation", "tests_are_truth"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "54031425e0e64b261b2d535a77ab6bd15b6a7473"}
{"id": "train_20994", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "documentation", "auditability", "security_gates"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "54e134a00b8de70d94962bbeeef60003e717d248"}
{"id": "train_20995", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "auditability", "security_gates", "ci_integration"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8613941af692f1afd16d123425c6653ee1308b53"}
{"id": "train_20996", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "evaluation_metrics", "tests_are_truth", "security_gates"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0258f13ab4a752259d013dee6421ec789b8343f3"}
{"id": "train_20997", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "security_gates", "tooling", "reproducibility"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b93541b9e97a019e6887ba18d7094d5d1fe61ca4"}
{"id": "train_20998", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "evaluation_metrics", "governance"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bef3ec228a81e0a139d3f6c00ff46377f1a3da85"}
{"id": "train_20999", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "ci_integration", "documentation", "repo_scale_reasoning"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d8bf04dd956bbae7bf036c184e72681b636162e6"}
{"id": "train_21000", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["documentation", "governance", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5ba9a1d300bed3758be39b89b9b2312c343b516e"}
{"id": "train_21001", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "tooling", "evaluation_metrics", "security_gates"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2c99db5669b1592b572f35ee83591b1d31e1313e"}
{"id": "train_21002", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "evaluation_metrics", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bb3006328eae7b2f78bae436f47253e261decaa0"}
{"id": "train_21003", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["auditability", "documentation", "cost_latency_tradeoffs", "governance"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b29b7103fe131a8b15e4f54f12c8e7f1f22c5a35"}
{"id": "train_21004", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "ci_integration", "reproducibility", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e9827c06b35bd4161421a911040357fac51ca557"}
{"id": "train_21005", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "auditability", "tests_are_truth"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "895992cd899766db8973f3889171cc8dd5dd9772"}
{"id": "train_21006", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "cost_latency_tradeoffs", "ci_integration", "security_gates"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b36c4c254a7041db5d9943567a21f6f3608fa95e"}
{"id": "train_21007", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "evaluation_metrics", "ci_integration", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e457350b6cee52c900f188376580b4a012add9fd"}
{"id": "train_21008", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "cost_latency_tradeoffs", "reproducibility", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9399c5c78c849c1944c4c64faa64d158d6802d66"}
{"id": "train_21009", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "cost_latency_tradeoffs", "documentation", "tooling"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f711b09e61504412f9c32ce18950264ab67b8cd3"}
{"id": "train_21010", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "repo_scale_reasoning", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d9768452fbb74886295779a4a8f0ba320fd2aeea"}
{"id": "train_21011", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "evaluation_metrics", "auditability", "tooling"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8bf5e786a5c18d4b47b282b8960a4da1312c9929"}
{"id": "train_21012", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "ci_integration", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "20e7f57364324a0cf3e50c7b55b4a4881271d050"}
{"id": "train_21013", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "security_gates", "tooling", "documentation"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fd31edc6ef37f24809912ae4b93637d6a6c14eba"}
{"id": "train_21014", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "governance", "tooling", "tests_are_truth"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a6a7e65500adef77f9c285853e553a63fab5d8a9"}
{"id": "train_21015", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["auditability", "documentation", "reproducibility", "tests_are_truth"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "af1e5559c490a3b47c721305cbf71a3a066709ee"}
{"id": "train_21016", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "tests_are_truth", "security_gates", "ci_integration"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "60ca990b73055c4879b2ccbd11bebe2b5de09844"}
{"id": "train_21017", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "documentation", "governance", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "01118ef55812ff1bc9065e055d94af05fdfe6546"}
{"id": "train_21018", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["governance", "tooling", "ci_integration", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8d2b3e4b4ffa9c07f7af81ee9bd82730b08cd978"}
{"id": "train_21019", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["auditability", "reproducibility", "security_gates", "evaluation_metrics"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "12d2b374ee8933a714724a399ed300c596958ee5"}
{"id": "train_21020", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "tests_are_truth", "governance", "auditability"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d403b76750d8a1617547e54328780a693bf1a078"}
{"id": "train_21021", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "documentation", "evaluation_metrics", "auditability"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "eaeba75462e8538e3cf6e585da44c827f8504e62"}
{"id": "train_21022", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "auditability", "cost_latency_tradeoffs", "governance"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6bf819ba8871a82b9c0ea5738d0323a1cdfebe6c"}
{"id": "train_21023", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "governance", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c0cee180d601c2951f056633f30a4f527efb9f84"}
{"id": "train_21024", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "security_gates", "reproducibility", "documentation"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9aa68f3bb76746968f7a8b8dd792604463af561e"}
{"id": "train_21025", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "ci_integration", "auditability", "tests_are_truth"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2bba3fc4ae467e6b048b1f7d7ca6a1f4380c7255"}
{"id": "train_21026", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "documentation", "auditability"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "08021b03d2768e03e3f4a67b9abdac50997e87cd"}
{"id": "train_21027", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "ci_integration", "tests_are_truth", "auditability"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8804f586605f7ce8ff06c20c3c68502783ca8bb1"}
{"id": "train_21028", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "repo_scale_reasoning", "tooling", "auditability"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4b7aae3697fb1f003af4be4514136f24840822f3"}
{"id": "train_21029", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "tooling", "evaluation_metrics", "tests_are_truth"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f4d664150e4789794d493d3e7182321dffc1fcf7"}
{"id": "train_21030", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "documentation", "security_gates", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9a7185c9b7219fece76ac2e9a75fc487e718a13c"}
{"id": "train_21031", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "tests_are_truth", "ci_integration", "evaluation_metrics"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "077762be642420f7683c458f68bc884d3ce084e1"}
{"id": "train_21032", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "reproducibility", "tooling", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1f256f3e212f8f7bf9304f9266814799d690d177"}
{"id": "train_21033", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "documentation", "security_gates"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e4888a1ade2272b1a531f97cb9d1a325fa55cdda"}
{"id": "train_21034", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "documentation", "reproducibility", "auditability"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "92a5bc9f74e9370469a3ecc24ae3a459736e5791"}
{"id": "train_21035", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "documentation", "ci_integration", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2304dea39fc90a2d1f1019e48d3501fae3657a16"}
{"id": "train_21036", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "tooling", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "92a77c14d72843a238787ba51c64eb059684f951"}
{"id": "train_21037", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["governance", "security_gates", "evaluation_metrics", "reproducibility"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ae97ed8990e3e90248c611b9e4f7ab9c0fcebc88"}
{"id": "train_21038", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "evaluation_metrics", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7ac96022d3f2c105cc1f9911547b40b8f2c0a326"}
{"id": "train_21039", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "documentation", "auditability", "ci_integration"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4dda38b527bb3a65522a495a85a9139c33ac6044"}
{"id": "train_21040", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "ci_integration", "governance", "evaluation_metrics"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0a0e89a9ea1abe78f64d2cfc542f921ae8d57bd9"}
{"id": "train_21041", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "governance", "tooling", "repo_scale_reasoning"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "803a7238698e9f76b43cbdcfd6c83042e77e7ee4"}
{"id": "train_21042", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "governance", "reproducibility", "ci_integration"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "caeeb6734682d10361dd67949fc6382c74373aa1"}
{"id": "train_21043", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "governance", "tests_are_truth"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d58c8e51ad7de6c24cf8f4e347e48b27a782bde8"}
{"id": "train_21044", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["governance", "cost_latency_tradeoffs", "tests_are_truth", "auditability"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c8c8c5d9a08fc7215c842bf5647a5298247590aa"}
{"id": "train_21045", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "cost_latency_tradeoffs", "repo_scale_reasoning", "tooling"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "04d39d98bd060bdc6a799330bbae28dcf5093cdd"}
{"id": "train_21046", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "auditability", "repo_scale_reasoning", "reproducibility"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e8cd808037595dccdd39861869af3869edd0ed16"}
{"id": "train_21047", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "security_gates", "documentation", "evaluation_metrics"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f692e7be9925b729e092530858ce3f841e948592"}
{"id": "train_21048", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "auditability", "reproducibility"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8fd85ffb8e5150dc761286ea234fc6667377c7cf"}
{"id": "train_21049", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "documentation", "security_gates", "tooling"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b788162b3f1bfceb92c177943f6a67439824a2b6"}
{"id": "train_21050", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "governance", "tooling", "reproducibility"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c9b7361fcc9cb465a2eedaa66ab49a47836248c5"}
{"id": "train_21051", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "security_gates", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b0cca1f995efcd6694e0a1ceb22c16e071e64e7f"}
{"id": "train_21052", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "security_gates", "evaluation_metrics"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b62f9cf18b5ffb2182f8559ed1ba35eed72637a6"}
{"id": "train_21053", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "security_gates", "repo_scale_reasoning", "tooling"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "88c77271b0341c6d190f8f20015c7241c430654b"}
{"id": "train_21054", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "tests_are_truth", "reproducibility", "ci_integration"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "47205dcd54583344d1401cea39ad019c6367a5a1"}
{"id": "train_21055", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "reproducibility", "ci_integration", "documentation"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f4348e2ace761aaa5df8acf931d24fd2639d928b"}
{"id": "train_21056", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "security_gates", "documentation", "auditability"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cfae7e09df4e92acf2f6c1b6e1a034ad7514fc25"}
{"id": "train_21057", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "reproducibility", "tests_are_truth", "tooling"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f7fb38abfd82faac7f45c2132b9a5c333760c794"}
{"id": "train_21058", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "tooling", "repo_scale_reasoning", "security_gates"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b76a148e23237b63d2e9d6088429c9a20b7068c8"}
{"id": "train_21059", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "tests_are_truth", "security_gates"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f2b0a8cdf2b1d1e5c7eb04bcc0551c89602067f1"}
{"id": "train_21060", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "governance", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fc29b09b1d31d8c2a9d9572984245a648d8348f8"}
{"id": "train_21061", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "ci_integration", "auditability"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1904b71906484f8e69973c39baa98716b486fb3b"}
{"id": "train_21062", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "evaluation_metrics", "documentation", "tooling"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4241c9ec8490360fc8eb770c7121903f7e2cdb62"}
{"id": "train_21063", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "reproducibility", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f5bda5832f8a7529af123f2d95c16e326198571e"}
{"id": "train_21064", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "tooling", "documentation", "governance"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3a7691b7d9ee067e57a9abc78ecc306b236645ba"}
{"id": "train_21065", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "security_gates", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7db75282fc9df12cd5db6a75e0ad9eef7310b73d"}
{"id": "train_21066", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "auditability", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4ed6169c9cdd118daf8da64d47d46cfa8b7f13d6"}
{"id": "train_21067", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "repo_scale_reasoning", "security_gates", "reproducibility"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "10fc9770090acc4e4c53c4993955290c28858a51"}
{"id": "train_21068", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["governance", "evaluation_metrics", "repo_scale_reasoning", "documentation"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "421f67df90eb0ebfaf66e9429038a959963d1d7d"}
{"id": "train_21069", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "010eeee1b4195e0f9f09e072f90d08280db83a1f"}
{"id": "train_21070", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "security_gates", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bef3ec228a81e0a139d3f6c00ff46377f1a3da85"}
{"id": "train_21071", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "reproducibility", "documentation", "security_gates"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6f7b2f8730c8ce252fc697ff5f3aea7e498a08ea"}
{"id": "train_21072", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "cost_latency_tradeoffs", "tests_are_truth", "reproducibility"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7958aa65de35f0d6ba2b616725f3dfb8cdb1d5ed"}
{"id": "train_21073", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "documentation", "governance"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "117d5aa3bb53de119878b3ebc8efdf8727a4b2d4"}
{"id": "train_21074", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "tests_are_truth", "cost_latency_tradeoffs", "governance"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d26e3d4cd4d201e949d9396e476c26f611293264"}
{"id": "train_21075", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "repo_scale_reasoning", "security_gates", "tests_are_truth"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0af844fc9d8fbcd96418ffa693816155b488f164"}
{"id": "train_21076", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "repo_scale_reasoning", "evaluation_metrics", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b49f4f98103efe58ef1d3d8e3471bf8b1c550e9f"}
{"id": "train_21077", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "documentation", "auditability", "tests_are_truth"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b9c5dca6bc639927cb35b65aaa14ed292641cd10"}
{"id": "train_21078", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "tooling", "cost_latency_tradeoffs", "security_gates"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "83a6e8c8ad16b50f012564f4ae0130fc4673433d"}
{"id": "train_21079", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "tests_are_truth", "documentation"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "59647fb77e1db8ebf504a56fea144566536fb4bc"}
{"id": "train_21080", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "repo_scale_reasoning", "documentation"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "89f7fb12f2e22ce6a33c1af98e73b6939bce68b9"}
{"id": "train_21081", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "governance", "tests_are_truth", "tooling"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1e85b8008fb78fa3ba0bd6a5431bcc5b6027c08a"}
{"id": "train_21082", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "documentation", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ac2d0f906a277b80239bdacbe4248c6b3b80056e"}
{"id": "train_21083", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "evaluation_metrics", "documentation", "security_gates"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "80c9cd241f1e3e86653e5dd2f1599ecc7a47b1b2"}
{"id": "train_21084", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "reproducibility", "evaluation_metrics", "governance"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "37bff38f43bfe58403adce4f8d2a6cda06b7c1e2"}
{"id": "train_21085", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "tooling", "security_gates", "ci_integration"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d22c282888083e14cc17e8499cc90d2b3137822e"}
{"id": "train_21086", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "tests_are_truth", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "14d4813a8238321d56dbc44e2efff303f8192991"}
{"id": "train_21087", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "documentation", "evaluation_metrics", "reproducibility"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "199b600466492d492eeec002eec48ee17d10d342"}
{"id": "train_21088", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["governance", "auditability", "security_gates", "ci_integration"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "aa89f9031ed37e4fd5ede39d21d4c4e5cb011a56"}
{"id": "train_21089", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "tooling", "governance", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e506a08a0c7ccfe55ab320de9c75d2fb2c25f333"}
{"id": "train_21090", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "reproducibility", "governance"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7958aa65de35f0d6ba2b616725f3dfb8cdb1d5ed"}
{"id": "train_21091", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "evaluation_metrics", "cost_latency_tradeoffs", "repo_scale_reasoning"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "16d363ff757c156e3f93487980f1de1b997f5e54"}
{"id": "train_21092", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "reproducibility", "security_gates"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5a5c21dce1a2ed0d97bce98c3ea416adea9d3b78"}
{"id": "train_21093", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "auditability", "tests_are_truth", "ci_integration"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "afa0caac479b01969d89d5a9a0049104999f06a5"}
{"id": "train_21094", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "evaluation_metrics", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1fe4259e9d32735b9e09a47a42897b2fa97e271c"}
{"id": "train_21095", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "governance", "documentation", "tooling"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fbaf1991db51564d4147aa26fdf65fcd28e613be"}
{"id": "train_21096", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "auditability", "tests_are_truth"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "617d91b037191add1855ce9f42b1a746f1cdb2e1"}
{"id": "train_21097", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "evaluation_metrics", "tooling", "documentation"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b630594deae0d9e4e500df3ec1d2568a4445cd2e"}
{"id": "train_21098", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "tooling", "documentation", "security_gates"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bfb0ae90860c5a0abc4c5d526bdac730214cd09b"}
{"id": "train_21099", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "documentation", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "006bb37b57654e404c47841a60080bee06d3c9f3"}
{"id": "train_21100", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "tooling", "auditability", "governance"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "79ccfde4615ec7e7e46f53e34ba696e8a7c498e1"}
{"id": "train_21101", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "auditability", "reproducibility", "security_gates"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1def3ad3c4bcfadccb26686a6fc266074f8ae666"}
{"id": "train_21102", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "ci_integration", "reproducibility"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4e20a757be83fcb99e60d83c845c084a5b3401c7"}
{"id": "train_21103", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "evaluation_metrics", "governance", "tests_are_truth"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "633c8fed64b1d09e5703d5bd7c27d7860d037c65"}
{"id": "train_21104", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "evaluation_metrics", "security_gates", "tooling"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f847235f55a1f8af2b755684037156d18f65be03"}
{"id": "train_21105", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "governance", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "238763008568fa4f434c1b1ad567686d29613abf"}
{"id": "train_21106", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "security_gates", "evaluation_metrics", "governance"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1c6ebe922cb6cd7da98b71d9663b2a8731e7af26"}
{"id": "train_21107", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "security_gates", "cost_latency_tradeoffs", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b1b2bbcdfea587c60692ec065d335bb9f0d5042a"}
{"id": "train_21108", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "documentation", "tooling", "auditability"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bb5d9073f0d6a6ad8db34a9ebc0ef2356d89ccd7"}
{"id": "train_21109", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "documentation", "auditability", "tooling"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "91b8f0d135ea3986fbd91ab44ffe1e26ef82a0b8"}
{"id": "train_21110", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "tooling", "evaluation_metrics", "auditability"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "eb6c32a23812fbf988694e83141ddfb53a4d3761"}
{"id": "train_21111", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "auditability", "governance"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d41742ac6bc3fee5ab93cd19f10ca0d2cf94c3ae"}
{"id": "train_21112", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "governance", "evaluation_metrics", "auditability"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5802a0f98c087a0c20648ae2ed809b286c4ee5f1"}
{"id": "train_21113", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "security_gates", "governance"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "28cf5859e4c5db33d862cad35d228a1f4efb5d52"}
{"id": "train_21114", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["auditability", "tooling", "ci_integration", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1a8b17b1cb74705fb2940a0c8fb16a70a73d2581"}
{"id": "train_21115", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "tests_are_truth", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fa90a1d70ea167148b68f9b25696b6851a09085f"}
{"id": "train_21116", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "governance", "tooling", "auditability"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5522f19133bd7f724f46a227fc1dbf3efe5ddab0"}
{"id": "train_21117", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "governance", "evaluation_metrics"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d6846185d3b6d7fbcb0a5a7b498e0c1f6c0424fd"}
{"id": "train_21118", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["documentation", "evaluation_metrics", "governance", "ci_integration"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "05c6cb390ef6183649ff613ee23a0979918ec671"}
{"id": "train_21119", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "governance", "evaluation_metrics", "security_gates"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a7bf04507d7b3c1fceb75eeb83bbe70621e92364"}
{"id": "train_21120", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "tooling", "governance", "auditability"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "889a51edcb34e0656885ea0451eec0f6343c0b42"}
{"id": "train_21121", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "security_gates", "documentation"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8a507252d7726b73a6a9ebc8afdaee601d0064e1"}
{"id": "train_21122", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "security_gates", "reproducibility", "governance"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "42e963f016de6732bb45009e463d598cf66389ce"}
{"id": "train_21123", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "documentation", "tooling", "security_gates"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5f73596c29bf88e946d2cd87fd926462cf3d5c21"}
{"id": "train_21124", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "documentation", "ci_integration", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "607ae216a930e460bbdd193d2f94b37bb2077e15"}
{"id": "train_21125", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["governance", "cost_latency_tradeoffs", "security_gates", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2c77e90072d0f5d804614fd2cad903d1abf479a7"}
{"id": "train_21126", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "ci_integration", "documentation"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8a2b49d7d04c4f2b0a7842e1d4cbefb167414090"}
{"id": "train_21127", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "tooling", "reproducibility", "tests_are_truth"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a25229f17f4afb874eb97fc3940c2a9a06b86683"}
{"id": "train_21128", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "tests_are_truth", "ci_integration", "tooling"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "59e23404751c9149fcc8af5cf103ee695e8ab37b"}
{"id": "train_21129", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "governance", "evaluation_metrics", "security_gates"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f83524357083e442397b2985e3b3e3657b19ea65"}
{"id": "train_21130", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "security_gates", "governance", "auditability"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6fd705d82a35bf58f21c591b03c88a0734c13576"}
{"id": "train_21131", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "auditability", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "026ca1c8ee3cea588df4dbd1cb8cc25e13743f76"}
{"id": "train_21132", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "documentation", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "900b16aed42d4bfe3f27c611a9757a2dafebd7e1"}
{"id": "train_21133", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "documentation", "tooling", "auditability"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b9e1b652ca48644fa41a440cfc7aa16573f9c3a9"}
{"id": "train_21134", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "evaluation_metrics", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3af89e68894b63a4f714a9df6d27e8dee8bca8ea"}
{"id": "train_21135", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "reproducibility", "repo_scale_reasoning", "security_gates"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6d14256ee5cde8c8d1b684b8544a3131ecddc95f"}
{"id": "train_21136", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "repo_scale_reasoning", "security_gates", "ci_integration"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1b8a68184df6741f884e413fc7cac30635e64351"}
{"id": "train_21137", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "documentation", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d1f00e37780ddaccaafd78bcc3ced7f46d1bf072"}
{"id": "train_21138", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "auditability", "documentation"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2f85335ab5fc955790e3dd944a5c28431af0434d"}
{"id": "train_21139", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "security_gates", "ci_integration", "tooling"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7993b00587c4c47639093db1e8fe2211b8f88c06"}
{"id": "train_21140", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "security_gates", "auditability", "governance"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1347f1d0cc438eb49879572fba6701cfe8537172"}
{"id": "train_21141", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "reproducibility", "governance", "documentation"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "86725fd5d0b7c729f2c8ea630f0e65bc2b396d96"}
{"id": "train_21142", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "security_gates", "governance", "ci_integration"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "787a4f2e5ef65c6ec6591dd56f9e3eed1288295e"}
{"id": "train_21143", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "governance", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3e7034ec50e3f3168dd31faff7f6ebf954c91080"}
{"id": "train_21144", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "documentation", "security_gates", "tooling"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cf442aa59af6370d0997f951a675ac717f125407"}
{"id": "train_21145", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "governance", "evaluation_metrics", "reproducibility"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "eb6ad5641d7fb5f8ca5ba13565ffd3f35ca776bb"}
{"id": "train_21146", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "documentation", "auditability", "security_gates"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cbf4a4f463c5a7b3539328f37105ca6a4e6ec51c"}
{"id": "train_21147", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "auditability", "governance", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8a76ce1daa46494db5f97be814f0c626fbc7c198"}
{"id": "train_21148", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "evaluation_metrics", "tooling"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dea06121061d2becaf37f82af694a1873c763e96"}
{"id": "train_21149", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "tests_are_truth", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "51d16a5ff59ec317bb43d0864ad62bd3f8604447"}
{"id": "train_21150", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "reproducibility", "cost_latency_tradeoffs", "security_gates"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1bcbba8e2e1ed8d60a5a9ba05e379729eabc731d"}
{"id": "train_21151", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "evaluation_metrics", "ci_integration", "governance"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "da662655b9a2ce01ac448c1682d3084eac374130"}
{"id": "train_21152", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "repo_scale_reasoning", "reproducibility", "tests_are_truth"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "57c38ec0d62d59d82d625c84f9355b7de20f46d8"}
{"id": "train_21153", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "tests_are_truth", "repo_scale_reasoning", "governance"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a2f67df1e60b9747a1fb43be9321e6abc5bebe0d"}
{"id": "train_21154", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["auditability", "governance", "tooling", "security_gates"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a20997bc106c7c38d6f00ced1a2986852d303ade"}
{"id": "train_21155", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "auditability", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "64b8049780989337f15324b95064371b31529467"}
{"id": "train_21156", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "governance", "tooling", "tests_are_truth"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1bd8e1903f1c4c9536ce980a3039513cf305bfe2"}
{"id": "train_21157", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "governance", "auditability", "ci_integration"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e2afc4e085baa518f62ea2f7d111b7498cb30531"}
{"id": "train_21158", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "tooling", "governance"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c622bc869fdeb177437f2491353e633cf9343858"}
{"id": "train_21159", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "reproducibility", "auditability", "governance"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "de37c771ebf8f6fcad5b461d88324de45e48eb7f"}
{"id": "train_21160", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "reproducibility", "auditability", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "148e5b8fb42d47bd234d72d6afa18a261f2f6cf8"}
{"id": "train_21161", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "auditability", "evaluation_metrics", "reproducibility"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8f185c85af35d945ee9d8d2d81e8563d8075d744"}
{"id": "train_21162", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "evaluation_metrics", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7c8f4e4aa0911dd7e269271d12e7550b3cb558f6"}
{"id": "train_21163", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "auditability", "tests_are_truth", "reproducibility"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "787e870cae58da2bca0cb25c6be09c8ac506be81"}
{"id": "train_21164", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "repo_scale_reasoning", "auditability", "tests_are_truth"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "697fe197b4c69042535501b398b7ffccad9bdb55"}
{"id": "train_21165", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "reproducibility", "tooling", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d222d41d852882743ec812e9c7ebcb21b1795ba2"}
{"id": "train_21166", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "governance", "tooling", "security_gates"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1b6cf117f3723a077de7855e5a35cb38985689f1"}
{"id": "train_21167", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "auditability", "tests_are_truth"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bbd9ad7491029c94ae1afccd9156c3dfe68e57aa"}
{"id": "train_21168", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "evaluation_metrics", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8258a50b8f1dae2ce1c77c38aad0305145c2f9fc"}
{"id": "train_21169", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "auditability", "tooling"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2042755836008b21fe280699f3911eda5aa503c1"}
{"id": "train_21170", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "tests_are_truth", "documentation"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "63fdd569e40d1df3d28437098c1ab046155f057c"}
{"id": "train_21171", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "evaluation_metrics", "repo_scale_reasoning", "tooling"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "276d9513d759a559bfd5d6745b8a11edb84bade6"}
{"id": "train_21172", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "evaluation_metrics", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dd4f3cbe9b4ffcc416a96aec372f8ddacdb2a1b1"}
{"id": "train_21173", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "tooling", "tests_are_truth", "security_gates"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7cb12ccfcec64edc2deae4ac26fd7bada0728751"}
{"id": "train_21174", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "tooling", "governance"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c70a6f44075c57b3df829065a346f7de60872c24"}
{"id": "train_21175", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "auditability", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e349a0114c3074380a490a98385a35cd19169bae"}
{"id": "train_21176", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "security_gates", "documentation", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "27e24dc5e095bb40ed90f66d42b17fdcd1331446"}
{"id": "train_21177", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "tooling", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3c48477b774ecd1e9497f3bc0511abeb3ccd439b"}
{"id": "train_21178", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "repo_scale_reasoning", "auditability", "documentation"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ae6e829973e052a5776f6a0c11c6ed7aef5d70df"}
{"id": "train_21179", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "documentation", "tests_are_truth", "security_gates"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "362b94273e81ff56047bb89fb520e4a8781ee5c1"}
{"id": "train_21180", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "governance", "evaluation_metrics", "ci_integration"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "86d42d17d273ea932fcb7c431041271fd1087387"}
{"id": "train_21181", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "tooling", "security_gates", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "29daefb43568b1fe794c1e57ee261c5289d6b1a0"}
{"id": "train_21182", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "tooling", "security_gates", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3a8257424400d932439794751744ee958501d299"}
{"id": "train_21183", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "ci_integration", "tooling", "governance"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ce26ac59da8e135d9b4bfcac0c18b1ccfcd56283"}
{"id": "train_21184", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "cost_latency_tradeoffs", "documentation", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7dd7e58ceb7b0b119730be82082089f353004754"}
{"id": "train_21185", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "auditability", "reproducibility", "tooling"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "daffb078574041307c1c0184c95f5de3d22abc07"}
{"id": "train_21186", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "auditability", "reproducibility"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7d5a6809087d9e93c337985babdce4e714e58c8f"}
{"id": "train_21187", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "tooling", "tests_are_truth", "governance"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "22b74eab73eb63c9c3fb8b12d2bf464725a2cbed"}
{"id": "train_21188", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "repo_scale_reasoning", "tooling", "documentation"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "097d847a608bf7df1143d7dc4870c3c99cadf746"}
{"id": "train_21189", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "ci_integration", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "40c9a1c539a8c33009e994aeeb15761943212b6c"}
{"id": "train_21190", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "tooling", "security_gates", "auditability"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7513d962fe005ba81a332ff7a1690233d043bc8a"}
{"id": "train_21191", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "auditability", "tooling", "security_gates"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e581359e4bd951f6010a9e53e9201e113708e47c"}
{"id": "train_21192", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "governance", "security_gates", "evaluation_metrics"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f7af38e29e43402c22c16098393cfc68c2582a37"}
{"id": "train_21193", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "tests_are_truth", "governance", "security_gates"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d04a14adb32b746aa16c12c416d393e9fec3515d"}
{"id": "train_21194", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "tooling", "ci_integration", "governance"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "94eece7310d16192f4b552e98556c8198b83682d"}
{"id": "train_21195", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "evaluation_metrics", "security_gates", "tooling"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0cbab1122d6b4ce7af40050b796a5cd54380dbb3"}
{"id": "train_21196", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "evaluation_metrics", "tooling", "tests_are_truth"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d8e84e5d69a67c1d19dd9c4a6326c31f48750948"}
{"id": "train_21197", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "documentation", "tests_are_truth", "reproducibility"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "52d6458540f025f765905f3e5db3dceef02f03b1"}
{"id": "train_21198", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "repo_scale_reasoning", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "29567d0500ab98f14a2958fac5ceb17363626003"}
{"id": "train_21199", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "documentation", "auditability", "tooling"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1b6043029b9bad6abf0b0f6becf5eb5bf218b862"}
{"id": "train_21200", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "security_gates", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b29b7103fe131a8b15e4f54f12c8e7f1f22c5a35"}
{"id": "train_21201", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "governance", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f413e73d7821a0d3e2557a3e4d8b029e21a7224e"}
{"id": "train_21202", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "ci_integration", "evaluation_metrics", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2fb16dd2ae8fc8333ae10570362e89b0c7962122"}
{"id": "train_21203", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "security_gates", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c9f8dbfa52a94a333c8050a16c705c3fe6460ccd"}
{"id": "train_21204", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "cost_latency_tradeoffs", "security_gates", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ac73df952c088e4ee7b542f6b70bc619dcca3076"}
{"id": "train_21205", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "evaluation_metrics", "auditability", "ci_integration"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dd2555e5d657c5f50830294b2a2594ed4c25ac32"}
{"id": "train_21206", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "evaluation_metrics", "repo_scale_reasoning", "security_gates"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "27ee7f625efb4a1c58c85f5d44b7cf43f99a5be4"}
{"id": "train_21207", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "reproducibility", "ci_integration", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "aef985048ee27a820a8a2856c0e9db98b9ded104"}
{"id": "train_21208", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "evaluation_metrics", "auditability"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "982bbcb56448c198d74919084d9174c8b443c84a"}
{"id": "train_21209", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "reproducibility", "tooling"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3b4bcf188c318c939d79b74c0cfec010f125f019"}
{"id": "train_21210", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "ci_integration", "documentation"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0e3bad6d7f29219c7e5c049f97e597ea6b825ec0"}
{"id": "train_21211", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fa4dc6e443f5e32a00745f5b3ef0891dc5bec644"}
{"id": "train_21212", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "security_gates", "auditability", "tooling"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4a522a4be4d779f4706df8ed9b94dcfce0ea1e1b"}
{"id": "train_21213", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "auditability", "reproducibility"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "60886d63ec783d4efd23b78c2e54e0821af58c52"}
{"id": "train_21214", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "ci_integration", "reproducibility"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cb1ded83e66eacd2de38957cd00f2c1fec0b2d2c"}
{"id": "train_21215", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "ci_integration", "security_gates", "documentation"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "97f95447c275faeb1c4b6c422d144b489904160c"}
{"id": "train_21216", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "auditability", "security_gates", "repo_scale_reasoning"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d8baea7632e20691eeb552623a6f929fac133beb"}
{"id": "train_21217", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "tests_are_truth", "tooling", "documentation"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dede9eba7d2283e8eb936d62b3fb96733cf8a97b"}
{"id": "train_21218", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "security_gates", "tests_are_truth"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a80dd3e3b3ea37e7a7139e55f06b67ea2c0adc9c"}
{"id": "train_21219", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["governance", "reproducibility", "security_gates", "tests_are_truth"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "23a221a47dc3d1fd3bb74c4d0fd848da69cd1d7f"}
{"id": "train_21220", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "repo_scale_reasoning", "documentation"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "448ae4ac8fd93f403e29d603c0ac27ad22d7bfe2"}
{"id": "train_21221", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "auditability", "security_gates"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4222a928f1fbf5767c20d5d94ccab828a38add54"}
{"id": "train_21222", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "auditability", "reproducibility", "governance"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5ce1d916a23344afe42a119efdebe396c2229b23"}
{"id": "train_21223", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "ci_integration", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "091c291929a221c120591429258dfbeec555ed28"}
{"id": "train_21224", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "documentation", "reproducibility", "tests_are_truth"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ce9fd6638891f486ca7a687fe426dccbbc48963d"}
{"id": "train_21225", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "reproducibility", "auditability", "documentation"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ea3ac5b3194d782f4dd229889537b9a50771175f"}
{"id": "train_21226", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "security_gates", "auditability"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ed9486d8f98390721558b628d802965c4de840a4"}
{"id": "train_21227", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "tooling", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1bb897aadc1e9059c721e8a82884ede593d577e8"}
{"id": "train_21228", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "tooling", "ci_integration", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "df7997378c247e89f04b04136740f59f01062413"}
{"id": "train_21229", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "auditability", "evaluation_metrics", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "583b43bc47de763b67ad40bd75a68b84005132c7"}
{"id": "train_21230", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["governance", "cost_latency_tradeoffs", "documentation", "evaluation_metrics"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e0b946571d7a368e38b02ec93d21af441a47959e"}
{"id": "train_21231", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "governance", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "699f7fb3b77e989c6d3fb751243333bc52cff1e9"}
{"id": "train_21232", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "auditability", "repo_scale_reasoning", "security_gates"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6c445b07747f4052d15bc3fa4fc7ed14ba62aef4"}
{"id": "train_21233", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "repo_scale_reasoning", "documentation", "auditability"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d6c024ed744735eb47e0a9aa8d821affe2a1c05a"}
{"id": "train_21234", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "repo_scale_reasoning", "documentation", "ci_integration"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b6227aa65ad22429c8f7c106587b9f2c0aa88908"}
{"id": "train_21235", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "repo_scale_reasoning", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8221472930450fb168f4136bd89732645971611e"}
{"id": "train_21236", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "reproducibility", "security_gates", "evaluation_metrics"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ef2327849b30a0c40e50fb2900cc7555338f69bf"}
{"id": "train_21237", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "governance", "tooling", "documentation"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f705a7b44f9f6ce2ee10ca837ad9b029ec86c883"}
{"id": "train_21238", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "governance", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f3a715e3150c1709eaaa20d73cd825a468e8f9d7"}
{"id": "train_21239", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "documentation", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1c44fb00b2fd77c2020b7097fcc440a310956456"}
{"id": "train_21240", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "security_gates", "evaluation_metrics", "tests_are_truth"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8b56f0c4f71c4c86a35a408e3080161e22dda34c"}
{"id": "train_21241", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "security_gates", "reproducibility"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "002ee14b4e180200260622a43d3285e67ae17169"}
{"id": "train_21242", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "tests_are_truth", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c4ffa8b3d0a41614b09e3ee915f76fe256a7d03f"}
{"id": "train_21243", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "evaluation_metrics", "auditability", "tests_are_truth"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "10b7d8bb507750a4d8004ed015f74eec2a76b24c"}
{"id": "train_21244", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "governance", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a72119e6e62ca0a9537d4ed358b5a8fedb5598b6"}
{"id": "train_21245", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "auditability", "documentation", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "334d0af8c2a27c5fd4b11bfcfffae66ca5977397"}
{"id": "train_21246", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "security_gates", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e8948e54f30c06b7b88d77e33af440b59adc57c2"}
{"id": "train_21247", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "documentation", "ci_integration", "governance"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9edc197a5c484c10e75cc190f95afee2868d12bd"}
{"id": "train_21248", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "tooling", "security_gates", "tests_are_truth"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "288371d5c3d08b43ed100d40932501bf4ed194ff"}
{"id": "train_21249", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "tooling", "security_gates", "reproducibility"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ff85c39f5fcce20411d75328e74d6611213a07b7"}
{"id": "train_21250", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "tooling", "documentation", "auditability"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3511b4cfabec8ca62ffe274c2263ed6788512c99"}
{"id": "train_21251", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "security_gates", "auditability", "ci_integration"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9cf546a5b0dab84c0c31f809fe1fcf48e28aa63d"}
{"id": "train_21252", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "tooling", "ci_integration", "documentation"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bb537470ffc6b12a9286d63b13e1c4228fa94b4e"}
{"id": "train_21253", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "governance", "repo_scale_reasoning", "tooling"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "94cef770bf3f93504adc346dc4bd734626d54a58"}
{"id": "train_21254", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "governance", "security_gates", "auditability"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "777a20f4cf4e4b4c212d7cf116e3cf382258502a"}
{"id": "train_21255", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "governance", "auditability", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ddd1e928c2f1433f03836f1a867b1d7990f7830c"}
{"id": "train_21256", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "ci_integration", "repo_scale_reasoning", "tooling"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "254b093138b9a947809e21767e40165814e8a13c"}
{"id": "train_21257", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "repo_scale_reasoning", "governance", "tooling"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9eee3dc5a5ae5b1aa8a15855c2703e59d754cffd"}
{"id": "train_21258", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "governance", "ci_integration"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "60defabc902705ae80599d207bc545dc18029852"}
{"id": "train_21259", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "security_gates", "tests_are_truth", "reproducibility"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2ee1e0b444da60df9208d10b361605b76cc573b9"}
{"id": "train_21260", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "evaluation_metrics", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d04e9d678d962445c9cef052cd0c58480c6ab1fd"}
{"id": "train_21261", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["governance", "ci_integration", "auditability", "documentation"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c2bcc87e69b9da422900554b9abd3ce39a0837d3"}
{"id": "train_21262", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["auditability", "reproducibility", "ci_integration", "security_gates"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b0e4054eb016d294c90b9a5ee2a24332db80fdcc"}
{"id": "train_21263", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "reproducibility", "tooling", "governance"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5dccc9285dbfc53981f4acaa9ba114ed056579f9"}
{"id": "train_21264", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "reproducibility", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "488a0b3281011bf8533f7ac504b32219d2da5bd4"}
{"id": "train_21265", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "reproducibility", "repo_scale_reasoning", "documentation"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9973b1a1002ffbc31af60fe0e8aa0acc9a66585b"}
{"id": "train_21266", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "repo_scale_reasoning", "cost_latency_tradeoffs", "governance"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "834d95543024ff4c0f73b1e2ebc6fd40b8d9d60f"}
{"id": "train_21267", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "governance", "repo_scale_reasoning", "security_gates"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3738671ead2fbfbdc65ab3ca601bb39977d512d6"}
{"id": "train_21268", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "governance", "repo_scale_reasoning", "reproducibility"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6197dd75b7938446ade53d5846aba8f1b28c74fe"}
{"id": "train_21269", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "tests_are_truth", "ci_integration", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a112c35499765c582d2e591467c83dce5398c648"}
{"id": "train_21270", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "documentation", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f46288e9a505f6766774f18988031329e49863e3"}
{"id": "train_21271", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "governance", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a6cc7d1316e21f57bd98c2ae395407eab27fcdc0"}
{"id": "train_21272", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "governance", "auditability", "security_gates"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7c571934f3cb0d8960f73b99b83630fab4430fd5"}
{"id": "train_21273", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "ci_integration", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b22cd9822ae6044e285f515e01a081dafcc99681"}
{"id": "train_21274", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "auditability", "evaluation_metrics", "security_gates"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b2caf81817ec3f6bd83b83952ed0197d402624aa"}
{"id": "train_21275", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "documentation", "governance", "ci_integration"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9b8727937fc2f7abb8533d971ba41ae703dd20f8"}
{"id": "train_21276", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "reproducibility", "ci_integration", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "28f130d870b1c14c8d0424f86c0577676c52b85d"}
{"id": "train_21277", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "reproducibility", "governance", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f1aa1aed0d18a52465b38436b6704a1385803d87"}
{"id": "train_21278", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "documentation", "ci_integration", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dc9b6219104e3a4ac12c8bb796663c2b09533ede"}
{"id": "train_21279", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "governance", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c400ebdba4e511ca70eba95f2423840c8b52423e"}
{"id": "train_21280", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "tooling", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "60defabc902705ae80599d207bc545dc18029852"}
{"id": "train_21281", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "repo_scale_reasoning", "evaluation_metrics", "documentation"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "62ed24d566c9e41f28b301f1980019f0eb77e0dd"}
{"id": "train_21282", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "ci_integration", "documentation", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0445b6e65a40198d41b897a1fa23d32a8b74d511"}
{"id": "train_21283", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "governance", "reproducibility", "documentation"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "18bc5779b56ba94f590e634511b34fc638e9e276"}
{"id": "train_21284", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "documentation", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bf87db71ac275f0a11264a686d137232e0129873"}
{"id": "train_21285", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "reproducibility", "repo_scale_reasoning", "tooling"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2d42b4786702d041cd49ba047be62892b9825f17"}
{"id": "train_21286", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "evaluation_metrics", "governance"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "57ff5dcf05c45aa47373bf7ba1711247b0e07ab6"}
{"id": "train_21287", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "auditability", "reproducibility", "ci_integration"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4b7652584b3a3e572b43871f92203736c465cdc4"}
{"id": "train_21288", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "tests_are_truth", "cost_latency_tradeoffs", "security_gates"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5f658bd54c134f24da9ded721e909b257f49da44"}
{"id": "train_21289", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "tooling", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f926d59708b3944153b23db9b9e1fba5fdd80552"}
{"id": "train_21290", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "auditability", "security_gates", "ci_integration"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "66df21a1d9bcf84eb0e808beae4cd5fba6e4a2c9"}
{"id": "train_21291", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "auditability", "evaluation_metrics", "ci_integration"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "127854640f009b147b84edbf373a38a71b5b2476"}
{"id": "train_21292", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "documentation", "ci_integration", "tooling"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0185ca9129cd5cb8fa9aec450bf9b66a3ad849e6"}
{"id": "train_21293", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "ci_integration", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8f9a3dcc2fa1d6fc0af38af9f6f8bf5ffa04400f"}
{"id": "train_21294", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "documentation", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a2c8ab97cd3717978fcd41dc2f6e6a0405c34e20"}
{"id": "train_21295", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "documentation", "tests_are_truth", "ci_integration"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fadfcbf18bc8b69a90944f43d6551da69cdec018"}
{"id": "train_21296", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "ci_integration", "tests_are_truth"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "278c9923152899982fe94ae7cb4461ab07958c25"}
{"id": "train_21297", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "tooling", "evaluation_metrics", "governance"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e9b580f79510c4f8d2a2c89322f985e435f624f2"}
{"id": "train_21298", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "cost_latency_tradeoffs", "ci_integration", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a418cf60f5fd718ef6aa33e8aefb7c73c96d172d"}
{"id": "train_21299", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "cost_latency_tradeoffs", "evaluation_metrics", "ci_integration"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c8b3f3a248e673f4000c6a9f2e5a9cdd95b367c1"}
{"id": "train_21300", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "reproducibility", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8db85154d0be67ff7a123d02aa0ed16a4dedb7e0"}
{"id": "train_21301", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "ci_integration", "documentation", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0596fea7e9bd2bdd7c952d2e10eb78f553df59c0"}
{"id": "train_21302", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "security_gates", "tests_are_truth", "reproducibility"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6e59758515628976fda184c0ee49273367b59b43"}
{"id": "train_21303", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "repo_scale_reasoning", "cost_latency_tradeoffs", "governance"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5295d582c15c6c091441dc9a86cbfb8ba93792ee"}
{"id": "train_21304", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["governance", "auditability", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4a715f6993037b22367662fd542125b0a960fce2"}
{"id": "train_21305", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["documentation", "security_gates", "reproducibility", "governance"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6c8ea11c84f2d343d15adb8d983ee5910c3cdfc2"}
{"id": "train_21306", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "governance", "repo_scale_reasoning", "documentation"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8ff94c75f34d6585bfc3fee6cd18bd881e7b894b"}
{"id": "train_21307", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "governance", "security_gates", "evaluation_metrics"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9d64e93cd58e6764bc9e7258f35e5c3c4119eff6"}
{"id": "train_21308", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "ci_integration", "repo_scale_reasoning", "security_gates"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ccd800a33f5b06ac7078cdd6187b5ba3b4d82f19"}
{"id": "train_21309", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "81128cd9f44b3f3de54059bbd89f575fd6006df3"}
{"id": "train_21310", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "security_gates", "auditability"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6a712f43dd1a65061cddd123a2865769f40b8346"}
{"id": "train_21311", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "repo_scale_reasoning", "tests_are_truth", "reproducibility"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9f2a12fd19c32440da93b0bc5276306fd38a7933"}
{"id": "train_21312", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "cost_latency_tradeoffs", "governance", "security_gates"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f9d8150b8ef0241640906ee7c316c7f3c9d62703"}
{"id": "train_21313", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "ci_integration", "security_gates", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "85c72fe1b1c6653efae1f1ce713def4acf7a8461"}
{"id": "train_21314", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "tooling", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4c047e725551bb29efc77cd10c918b7d1b48f6ca"}
{"id": "train_21315", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "governance", "security_gates", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ea1cf397984a67b7e8c19b38836236fb0831b6b5"}
{"id": "train_21316", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "documentation", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3902d32f6becdf32cd25b0f373f7ceffc4754372"}
{"id": "train_21317", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "governance", "ci_integration", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d91b3f2ce61f7c366d37ee695f39930b6d3ffe08"}
{"id": "train_21318", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "reproducibility", "security_gates"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e48dcf984b0fdc01dea49278cb5d7b04a006877a"}
{"id": "train_21319", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "repo_scale_reasoning", "governance", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "eb200d83f3a04e4d1986c4ee974a88aeb53e552f"}
{"id": "train_21320", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "ci_integration", "documentation", "governance"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b1e07f99e323faef5ab45485d4205974fc50b51d"}
{"id": "train_21321", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "documentation", "governance", "auditability"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c63d575bd4a1964abd89535bb821c3207a8da73f"}
{"id": "train_21322", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "ci_integration", "auditability"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a58e065cee175022e7bf9c4a9510235c517cd881"}
{"id": "train_21323", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "reproducibility", "cost_latency_tradeoffs", "security_gates"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0b6e8d09a10d7a0ea4ae68ee51651b1acbcadff0"}
{"id": "train_21324", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "governance", "auditability", "tooling"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9112cea917a2eb62a0bff9317fe639e7e972c703"}
{"id": "train_21325", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "cost_latency_tradeoffs", "reproducibility", "governance"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2b62068c5dd4774b81d4efa1d17ae73171923df3"}
{"id": "train_21326", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "security_gates", "evaluation_metrics", "tests_are_truth"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7e4bfd369eb6ebffa69543e3a9344b34a5cf8e3f"}
{"id": "train_21327", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "reproducibility", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "93d887997fa8e6356c0de7733f23a674ad42f8ab"}
{"id": "train_21328", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "auditability", "governance", "ci_integration"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8acd471390ed7b6737c35e325b548ebe7e4c98f2"}
{"id": "train_21329", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["governance", "evaluation_metrics", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e650a6101bb8dfacea37eb37f08a27a85a31b6ac"}
{"id": "train_21330", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "governance", "tests_are_truth", "auditability"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a62705be509bab230b406e90a4cfa81ad2492749"}
{"id": "train_21331", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "tests_are_truth", "auditability", "reproducibility"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "96afed11001035b188f840f98cdc74f508fa6caf"}
{"id": "train_21332", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "auditability", "documentation", "reproducibility"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b32953853fa9e80d0a998e3029349fe61a4fa595"}
{"id": "train_21333", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "ci_integration", "governance"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e70e764a5667f9546050efe912b3c0d7c795639e"}
{"id": "train_21334", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "documentation", "evaluation_metrics"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3b146308eaf0ec9248fd4508f86b5cebf5a5065f"}
{"id": "train_21335", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "documentation", "tooling", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c486e59809c4f0777e82b954d08994cefdd080e4"}
{"id": "train_21336", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "evaluation_metrics", "tests_are_truth", "governance"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "de33d39dee7b034c72d1a50669f1173718a2a60a"}
{"id": "train_21337", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "documentation", "security_gates"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "eebeb89bfaa6bfb01f6956937e2d8bb167bcd63a"}
{"id": "train_21338", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "evaluation_metrics", "documentation"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ce93b072bfa41e50ecff4980a1743ab07785c76f"}
{"id": "train_21339", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "tooling", "security_gates"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cab46df1ebd66e5af23356b386d7bd1c14db038c"}
{"id": "train_21340", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "security_gates", "reproducibility"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0a248b4e880499034014812c24165172a7bf4200"}
{"id": "train_21341", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "documentation", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b75e83242d6a8f024c166f7cbe9b9e1dbd693335"}
{"id": "train_21342", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "security_gates", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "982b08c182f3764beb5179a6841991851096b61b"}
{"id": "train_21343", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "ci_integration", "auditability", "security_gates"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "88aca33060672c0284b26896932c296e1a3a896b"}
{"id": "train_21344", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "ci_integration", "evaluation_metrics", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5a8711c515613be5095d04e2a5687e98b0b3dfa8"}
{"id": "train_21345", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "reproducibility", "repo_scale_reasoning", "evaluation_metrics"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7012427da08b93070916f45e327d179d47749983"}
{"id": "train_21346", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "tests_are_truth", "cost_latency_tradeoffs", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "93f562ed0b739f146dd447b35f13b47be9519366"}
{"id": "train_21347", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "auditability", "security_gates", "documentation"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e5ab96165991ebc7092e5d6dfd79d019616eaa15"}
{"id": "train_21348", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "documentation", "ci_integration", "security_gates"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "95d630d668d0debba33a1b618ee9f71731814562"}
{"id": "train_21349", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "ci_integration", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9187ebd929618d15a55f1a1799c262c83b4e549c"}
{"id": "train_21350", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "tooling", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "40693deb3c0ac8ca0d9b6e157863a3e1294270e7"}
{"id": "train_21351", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["auditability", "governance", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0b89cf6e876db9880f397da703cbb399924c386e"}
{"id": "train_21352", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "evaluation_metrics", "reproducibility", "security_gates"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ba07f3b02ff71b3cf4abe193794831124f782b68"}
{"id": "train_21353", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "auditability", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d1a8cf7889bac99ef7fe068b84e7a74c1682be13"}
{"id": "train_21354", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "repo_scale_reasoning", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2bdf5f454e32fb6f10b4b58e437135d728821527"}
{"id": "train_21355", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "tooling", "evaluation_metrics", "ci_integration"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "469a14d3a59d0f5d5202893c7a299b87aad54df7"}
{"id": "train_21356", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "repo_scale_reasoning", "ci_integration", "tests_are_truth"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d89fabf4e57f1adf45d091a12be00c9e23d70812"}
{"id": "train_21357", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "tooling", "ci_integration", "tests_are_truth"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4c5bf6a351139aa6aa673fc582159a972a4eced0"}
{"id": "train_21358", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "governance", "ci_integration", "reproducibility"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c3672d6eaee351254c3935b9cbd87db24b21ec1b"}
{"id": "train_21359", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "tests_are_truth", "repo_scale_reasoning", "documentation"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "727dd238e03a35d7233194c480eefdc10bf317c3"}
{"id": "train_21360", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "evaluation_metrics", "governance", "ci_integration"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2cf779481ea613fa8415d01ad9c4e2f156c0a232"}
{"id": "train_21361", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "documentation", "governance", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "05c9bbf913a8a84e86550e798a56e0846900282d"}
{"id": "train_21362", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "documentation", "auditability", "governance"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cc4ce30b838e090f53809a7350a0b1324b7ba1d5"}
{"id": "train_21363", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "auditability", "governance"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "034d2a3e6e9998d6ed59ea1252f09d2ffaa1dce7"}
{"id": "train_21364", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "repo_scale_reasoning", "security_gates", "ci_integration"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0f52002d38b4a78093254600e888a00006ae1292"}
{"id": "train_21365", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "tests_are_truth", "evaluation_metrics", "tooling"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4cb41868de41a4990a125354269d89cd3032edce"}
{"id": "train_21366", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "tests_are_truth", "tooling"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6379c691d63a438feebd21dcae5f237570633286"}
{"id": "train_21367", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "repo_scale_reasoning", "reproducibility", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e744c1e07b81c45fafc4f4b2650ac2f09d62c192"}
{"id": "train_21368", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["governance", "cost_latency_tradeoffs", "evaluation_metrics", "tests_are_truth"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8838b9665a90f082a348223bcb12c688bb946153"}
{"id": "train_21369", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["governance", "evaluation_metrics", "ci_integration", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "016a8df867549d7005d299c033ea5409b45c8de6"}
{"id": "train_21370", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "documentation", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "82d56bd5dbd36f9b154da297fd9a6dfffc743474"}
{"id": "train_21371", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "tooling", "reproducibility", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fd61db632de29ac4a806fb6c8ac15e29112acbc7"}
{"id": "train_21372", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "tooling", "reproducibility"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "96134be95f722789129fc500bab5a09ad25574a1"}
{"id": "train_21373", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "security_gates", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cb80ae6bf01f0bb786664e3a932cd0fd40651f04"}
{"id": "train_21374", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "documentation", "governance", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bd5f24cbda3fd1fdb5f3f62390be7732ab9145db"}
{"id": "train_21375", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "reproducibility", "tooling"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9099a2502621d5d6bafbd5df04cc86e4791c0f11"}
{"id": "train_21376", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "repo_scale_reasoning", "evaluation_metrics", "reproducibility"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f9951c185210cb6e53c6bac087c9c6089a8ef37c"}
{"id": "train_21377", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "tests_are_truth", "security_gates", "ci_integration"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2760483a3ba592fb2b88eec49cab627fba146e96"}
{"id": "train_21378", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "security_gates", "governance", "tests_are_truth"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bdb4665545b056d0f661047ed8a57d11a95f1665"}
{"id": "train_21379", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "governance", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "54e57c9a53644ba59b9e4fd596ba26b62dc955c0"}
{"id": "train_21380", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "documentation", "ci_integration"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2d11b26b0121c720bc444a61611bba3b5d638841"}
{"id": "train_21381", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "documentation", "tests_are_truth", "ci_integration"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "941bb3e6de7f4cb469d2b87858adb4a8a74e942b"}
{"id": "train_21382", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["governance", "tests_are_truth", "auditability", "reproducibility"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2dc3833ba7f87e8fd8db50be4907316aee5bd682"}
{"id": "train_21383", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "auditability", "security_gates"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4de4b65480836023245f92f5a13d76238d3c5b96"}
{"id": "train_21384", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "ci_integration", "documentation", "governance"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0fa2ce7c1e6552cdbf2e793d916dc12e66e44518"}
{"id": "train_21385", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a12f13f1f11e92859e6af45aa5c14c5fed079260"}
{"id": "train_21386", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "ci_integration", "tooling", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b0170cb245c2e86099f7a3757bc141f4ac3a8ea4"}
{"id": "train_21387", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "security_gates", "auditability", "reproducibility"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6efa90cea24d69bee5332cec1f8bcbce2d8d43d9"}
{"id": "train_21388", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "security_gates", "governance", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a7f536c8d99c8891b546a248bfe540e96af8a8e4"}
{"id": "train_21389", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "tooling", "ci_integration"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "093b3e4b909b4155ff03b6b447ebea47fc8b278a"}
{"id": "train_21390", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "security_gates", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "11523e259300fe87aafb75fad9a80bb2a744fc83"}
{"id": "train_21391", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "tests_are_truth", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ed50f9ca036d3dddc27f5dda796fe5179d3f272f"}
{"id": "train_21392", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "reproducibility", "governance", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6e3017df39fba547d97bd3c6668f954455f62d17"}
{"id": "train_21393", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "documentation", "governance", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "170f18959c15440e6eeec8b100d94897f41cef77"}
{"id": "train_21394", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "evaluation_metrics", "documentation", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5da81fdb6dbd0b29cc75f1b9f28143b06629ba9a"}
{"id": "train_21395", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "auditability", "evaluation_metrics", "governance"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "987f5e1f9d72becefdf43c0b2a1c85a29fe10025"}
{"id": "train_21396", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "documentation", "tests_are_truth", "reproducibility"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1c6ebe922cb6cd7da98b71d9663b2a8731e7af26"}
{"id": "train_21397", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "governance", "auditability", "ci_integration"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "612be51a66e8ba7a7aefcaf1162cbbe1969f8cd8"}
{"id": "train_21398", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "documentation", "tooling", "tests_are_truth"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "139f5bff4a5875e47d8261d8b9ad22eb7b557e98"}
{"id": "train_21399", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "evaluation_metrics", "ci_integration", "auditability"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9f57d2e8d5800e03bf2763d00f6c3803e87bf6de"}
{"id": "train_21400", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "auditability", "ci_integration", "tooling"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a19095ce24a4610417aba3096103c4416fb0dfa3"}
{"id": "train_21401", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "evaluation_metrics", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4fb14106fb529082303f2d5c6f70e1559cdbe21c"}
{"id": "train_21402", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "tests_are_truth", "security_gates", "ci_integration"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c3ba53d1786a4904c3cfd9b889d2058fff712ee4"}
{"id": "train_21403", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "evaluation_metrics", "governance", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "750886e5b3c16dcda42ed6086f7a2a7cc968ba69"}
{"id": "train_21404", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "documentation", "repo_scale_reasoning", "governance"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c99cd8c4d9bc16107b487643bc12cbf1fb2e5599"}
{"id": "train_21405", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["governance", "repo_scale_reasoning", "ci_integration", "tests_are_truth"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4d25a3c21c865391fe7f4ae1de28c11a4b8d4458"}
{"id": "train_21406", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "governance", "reproducibility", "tests_are_truth"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c04d542851ade3e87394d290bd1947871097a346"}
{"id": "train_21407", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "reproducibility", "cost_latency_tradeoffs", "security_gates"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7cbda16d4faa66659e0b6161f88a6eb9cf7895eb"}
{"id": "train_21408", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "evaluation_metrics", "documentation", "governance"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0118ade6aa42a4a569832f1426720bd1f8672ca5"}
{"id": "train_21409", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["documentation", "tooling", "ci_integration", "governance"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5f9b927e0bcf7a8a815ae327449ae36d2812ef58"}
{"id": "train_21410", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "reproducibility", "repo_scale_reasoning", "tooling"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "91e0b0fc203b5297366a0b8eea1728f2ff48b3fa"}
{"id": "train_21411", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "cost_latency_tradeoffs", "auditability", "evaluation_metrics"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "13ba0ed0d6ba2b53cb33b1e54b295dc0218e9d36"}
{"id": "train_21412", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "reproducibility", "tests_are_truth", "auditability"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b9088aafa87b7814a754d27d020f7254e42d0864"}
{"id": "train_21413", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "governance", "repo_scale_reasoning", "security_gates"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1fbbedef2e55f3fa5b8d85b7b8eb97af0289c53f"}
{"id": "train_21414", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "documentation", "ci_integration", "evaluation_metrics"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "87361510ee0d276453712ae0171d9c640cbb29e9"}
{"id": "train_21415", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "tests_are_truth", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a06f9876030402d86d79e1cff4820b19618bba97"}
{"id": "train_21416", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "governance", "documentation", "auditability"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "85f992950c131dd1392134f717c80ba904d7e0d2"}
{"id": "train_21417", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "evaluation_metrics", "governance", "ci_integration"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1b5188e1c26b9beee427baaf79b5ae5d9c702b81"}
{"id": "train_21418", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "documentation", "tests_are_truth", "tooling"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "290f44af830f8f6c86ff1d0d38f53c034514bab4"}
{"id": "train_21419", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "auditability", "evaluation_metrics", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "315c902e1abdbcd56e0c3d53d4e81ec9314f9e40"}
{"id": "train_21420", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "reproducibility", "ci_integration"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7bf9c5e01d2b1c03bac51612de19a7767d921537"}
{"id": "train_21421", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "tests_are_truth", "governance", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cee6cb588d4a3c2dca32401f5ac74dbc624198ff"}
{"id": "train_21422", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["governance", "ci_integration", "security_gates", "evaluation_metrics"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "92994400e6e5d57b097b963e60cbb9a811ecd520"}
{"id": "train_21423", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "cost_latency_tradeoffs", "security_gates"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "16161d72e2f69a8f60e9ceb0ccda61748b77a31c"}
{"id": "train_21424", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "governance", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7e59645475f3ed9608d6fe900b90dc0bfbfefd90"}
{"id": "train_21425", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "security_gates", "tooling"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fd5e21b37e1f9a6c49d62caa0ef0a4a6ddad2191"}
{"id": "train_21426", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "cost_latency_tradeoffs", "reproducibility", "tooling"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1c4c1beb526a8c6ee36eddb1daffa3e309dbfcf8"}
{"id": "train_21427", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "auditability", "tests_are_truth", "security_gates"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e523bd2a9474230c0e30815f4210eb0d9709c656"}
{"id": "train_21428", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "auditability", "evaluation_metrics", "tooling"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2e3fae20c30ebd2c974c6b00365c6ce4fbd882c4"}
{"id": "train_21429", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "security_gates", "ci_integration", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c64bb68a97dc5c53d9f44faf55ee1773a9099c3e"}
{"id": "train_21430", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "reproducibility", "documentation", "governance"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3d09cec57ae79c362a9986882402146d1858b5cc"}
{"id": "train_21431", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "auditability", "documentation", "reproducibility"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0a6b0c1af17a963f055446e98d553d05a400b03a"}
{"id": "train_21432", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "evaluation_metrics", "documentation", "tooling"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "da9c6033cc5725cb5c7934c92aadc2091221ca3c"}
{"id": "train_21433", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "auditability", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "26d903020975cf29a13d1cf4a85e90da1ed159c9"}
{"id": "train_21434", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "security_gates", "reproducibility", "auditability"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a81636d0843081cff181ef2f886eeb812823bd18"}
{"id": "train_21435", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "auditability", "reproducibility", "governance"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9b8727937fc2f7abb8533d971ba41ae703dd20f8"}
{"id": "train_21436", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "evaluation_metrics", "auditability"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3f962232e2a76e250c576a1fdd9c72bd72337550"}
{"id": "train_21437", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "reproducibility", "auditability", "security_gates"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f00cf79c44ddaac960bf6028727cd4cedfe31405"}
{"id": "train_21438", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "ci_integration", "repo_scale_reasoning", "security_gates"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bc30d6c375eb1a84459f338714e110b18bc8aaa3"}
{"id": "train_21439", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "ci_integration", "governance", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "67f141a4724be2f5964f9b8e6016ef70ff96964f"}
{"id": "train_21440", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "auditability", "repo_scale_reasoning", "documentation"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a487c19341c1d0a1bf9f961e99137a30cd73ff88"}
{"id": "train_21441", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "evaluation_metrics", "auditability", "tooling"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "eabf523762507f8b36a984c718070334693d41d1"}
{"id": "train_21442", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "tests_are_truth", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6e21dc9bee992a3d23a46aca0127d9d9a2255842"}
{"id": "train_21443", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "auditability", "tooling"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a61c7f734a478d547cbd9a72e91341f411d9570b"}
{"id": "train_21444", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "documentation", "ci_integration"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4f300847821fedb7359be8dc3ec563e23df4dd72"}
{"id": "train_21445", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "auditability", "governance", "tooling"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f4c87184625c76a4678cdd7b9196b80d1f9e6708"}
{"id": "train_21446", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "ci_integration", "repo_scale_reasoning", "governance"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "df52108b142d67fa025de349cb6a28a6f9b84526"}
{"id": "train_21447", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "evaluation_metrics", "ci_integration", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3a6b83fac7e9ad52a4d97a6f53f082ec7301c9c8"}
{"id": "train_21448", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "governance", "security_gates"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "81e427f1397311547e4783d9d41d470779697a61"}
{"id": "train_21449", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "auditability", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "618af62be595877672fc471546157aff8165d4bf"}
{"id": "train_21450", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "tooling", "ci_integration"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "97a3a899314ae5561eb57a5345fe5253488f20a3"}
{"id": "train_21451", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "security_gates", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "173fd23a9efe25d66a612ee863f2fc613f2cf10c"}
{"id": "train_21452", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "41103ab0eaedf05277d1ded0c74a4c8d5dd95461"}
{"id": "train_21453", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "documentation", "reproducibility", "auditability"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "918d04f3249f714350cd368d578da59d6f41ff1c"}
{"id": "train_21454", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "documentation", "reproducibility", "auditability"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2fd0518d5399114fa0885ffdfa27bff2d663307c"}
{"id": "train_21455", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "tooling", "auditability"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f00e9cf33dbd9c56e30b63b2f171f1df7607ab77"}
{"id": "train_21456", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "ci_integration", "security_gates"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8052f26f2c0cc4af741aabc6dc902644cfb04e45"}
{"id": "train_21457", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "documentation", "governance", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e1a2a0fa91a568c9ca4c7481fed27d5915dadc49"}
{"id": "train_21458", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "cost_latency_tradeoffs", "documentation", "security_gates"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "70358dbd5cb00f3f6e6cc90c86485d6d42c6db6c"}
{"id": "train_21459", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "documentation", "evaluation_metrics", "governance"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e67a2688833899a76c5502a009a4bcb132a868c4"}
{"id": "train_21460", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "reproducibility", "documentation", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "eed67ba6cb3af11e05409f18b290cc43f95c74be"}
{"id": "train_21461", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "security_gates", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cf18e40907b3d78787ba2c6dbb7119e77ec8a3f6"}
{"id": "train_21462", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "tooling", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a50b4d8d1dcf4e3cfc90b0663ae665ebbc6823fa"}
{"id": "train_21463", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "security_gates", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c98c1443db7fdaae2956f01687510bd09354de42"}
{"id": "train_21464", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "documentation", "tooling", "governance"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "548367debbff24708a28e56e87b1be2cf928bdac"}
{"id": "train_21465", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "ci_integration", "evaluation_metrics", "governance"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f40d1ba66d2ded3683fec1b6e9ace27fb12674a6"}
{"id": "train_21466", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "tooling", "governance", "tests_are_truth"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7db75282fc9df12cd5db6a75e0ad9eef7310b73d"}
{"id": "train_21467", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "governance", "ci_integration", "reproducibility"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6868d4c8dcfaee0d017ab1f2a7bb71f20adc0bb8"}
{"id": "train_21468", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["governance", "evaluation_metrics", "documentation", "tests_are_truth"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "205b10828e58b0c7b28dca5397e808f39b13149e"}
{"id": "train_21469", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "security_gates", "repo_scale_reasoning", "governance"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dc171872647a540aa3dec5f515f9887ec5a6bd58"}
{"id": "train_21470", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "tooling", "repo_scale_reasoning", "governance"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "53014c05562c37fb24910a9532b3fa92e72a263a"}
{"id": "train_21471", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "auditability", "ci_integration", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "08eaad429878901867a4ed84394614fdab084517"}
{"id": "train_21472", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "auditability", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e349a0114c3074380a490a98385a35cd19169bae"}
{"id": "train_21473", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "documentation", "reproducibility", "ci_integration"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5e1422b3d061bbf3e4ffdc22e4da441c1b768e45"}
{"id": "train_21474", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "repo_scale_reasoning", "reproducibility", "governance"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d9ee9d9b1607af9c99a62a6bc1630e4ba79d6c31"}
{"id": "train_21475", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "reproducibility", "evaluation_metrics", "auditability"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "22d0e1c78c2e6cacaa2ff8fd1e6a7bb4b8973d70"}
{"id": "train_21476", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "reproducibility", "documentation", "tests_are_truth"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2cca50e8258a6c95434b8214df77395d44390b71"}
{"id": "train_21477", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "tests_are_truth", "tooling", "documentation"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9d6d756e0315e5d9bd853ce39b11da6d41fde469"}
{"id": "train_21478", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "governance", "evaluation_metrics", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5547391c52f3c79f1142268fa0b606dcf01210ad"}
{"id": "train_21479", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "security_gates", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4bf7033557afbd9bae7a60b76b9b0bd65c825618"}
{"id": "train_21480", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "auditability", "tooling"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "24760afb3305041c9082682c4985f1774256cb41"}
{"id": "train_21481", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "ci_integration", "tooling", "governance"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f7cdb14a348e63ea72def56d9352a4dfe9039a83"}
{"id": "train_21482", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "security_gates", "reproducibility", "auditability"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9153b2c4b98e428ab11a8175a3b01f5cb74c6d92"}
{"id": "train_21483", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "documentation", "governance", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e38a560e3cc2cade7c62ee014f456d67df51386f"}
{"id": "train_21484", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "tests_are_truth", "repo_scale_reasoning", "governance"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cc22772d748d6393692d0d6e40663f6c0d731d55"}
{"id": "train_21485", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "tooling", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a4f826c99e851c51112aa8ae83d81b0c5cd695c0"}
{"id": "train_21486", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "governance", "reproducibility", "auditability"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0923183260a00f8e8323a216a2f386119955e7d1"}
{"id": "train_21487", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "repo_scale_reasoning", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4c3823c8228dc2395ec6a57f2cd570f00f5500c6"}
{"id": "train_21488", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "reproducibility", "governance", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "004208a7bdfe34ecc35bc8510d583daeb5fba804"}
{"id": "train_21489", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "documentation", "tests_are_truth"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "138c63d8290f16c8d3a3a3ff8093759e3762ac50"}
{"id": "train_21490", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "reproducibility", "repo_scale_reasoning", "security_gates"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "74286fd36a3aa32e97351226f5c40afe4c0709c5"}
{"id": "train_21491", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "security_gates", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d131bb3522c1ff551a7bf3ec74d60337b8092183"}
{"id": "train_21492", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "reproducibility", "auditability", "security_gates"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "94124218079618e89bcf13b545a0c6568cb8fed5"}
{"id": "train_21493", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "repo_scale_reasoning", "documentation", "governance"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "efec2bc38ed2825ab033b00e306e5e60abb76315"}
{"id": "train_21494", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "tooling", "security_gates", "ci_integration"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4ad8499d8a36d08b32bde056c894f38b6851e8a8"}
{"id": "train_21495", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "tooling", "ci_integration", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6769bcb215ee73c3d4d46905f01510aa9abc695a"}
{"id": "train_21496", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["governance", "tooling", "repo_scale_reasoning", "reproducibility"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a1df41053910a47af820e361af83a4fcfd02cf40"}
{"id": "train_21497", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "ci_integration", "governance", "tests_are_truth"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1a631f3cb29e95fc1dd9917f58f69ceb7b88b31f"}
{"id": "train_21498", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "auditability", "documentation", "tooling"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2f35d5e99fa3e6dee811487a7c3949f43e961516"}
{"id": "train_21499", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "auditability", "tooling"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "34cfe6bdbfb41e5f99a6b5405f311c2976d84221"}
{"id": "train_21500", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "ci_integration", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c15c1216978dfea0fc55f49a0aa7aa20b26c1943"}
{"id": "train_21501", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "tests_are_truth", "evaluation_metrics", "documentation"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c7f191d25046b1fb00634d519038e012cc7fedde"}
{"id": "train_21502", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "security_gates", "documentation", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "33ce75fd7291f58f4ce6dca5bfa53af4be458c42"}
{"id": "train_21503", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "ci_integration", "security_gates", "evaluation_metrics"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e0509ed9c7a40ea181c881adf2cc3bb69ffe20cc"}
{"id": "train_21504", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "evaluation_metrics", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6d14256ee5cde8c8d1b684b8544a3131ecddc95f"}
{"id": "train_21505", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "repo_scale_reasoning", "tooling", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "09bd3afff29dd4fd083a13a211e23a0db71974b3"}
{"id": "train_21506", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "tooling", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "705f1e88af4a51ebb669e9f749f72898ed3e5220"}
{"id": "train_21507", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "reproducibility", "tests_are_truth", "governance"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "90dd2fc3f9709be06e8a0c55531174e8df91ac97"}
{"id": "train_21508", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["governance", "tooling", "security_gates", "auditability"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f54dad4831a2058bf6a15183187fc24987b7fab4"}
{"id": "train_21509", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "security_gates", "reproducibility", "tooling"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b08c51b11632a1a652fd68be6f1adbcc7d53d702"}
{"id": "train_21510", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "security_gates", "ci_integration", "reproducibility"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "25ff2e8aa40359a4488e477e34e8b2fc5742636b"}
{"id": "train_21511", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "cost_latency_tradeoffs", "documentation", "governance"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bcbb7373a2a6bcf36ddfe19ff331aed0c7162bdc"}
{"id": "train_21512", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "auditability", "reproducibility", "governance"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "31bb34a20aae8a557c9cb451f4d20abe55c3ec6d"}
{"id": "train_21513", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "governance", "tests_are_truth", "auditability"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e80d123ec44434d28c3e90be240b8b17ca9d0e42"}
{"id": "train_21514", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "reproducibility", "tests_are_truth", "governance"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "65d6c1a681f3da4e6dcccf4ab7f8a04c7acc97b2"}
{"id": "train_21515", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "security_gates", "ci_integration", "tests_are_truth"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "78dd12677defa01d706b5affbfc7b810912f68a3"}
{"id": "train_21516", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "reproducibility", "ci_integration", "governance"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "451a7469450697924695b23a24c1b8ec9b582b63"}
{"id": "train_21517", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "tests_are_truth", "tooling", "auditability"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "29deb8c97d43bb45e089845c665edb80813710f5"}
{"id": "train_21518", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "documentation", "ci_integration", "tests_are_truth"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d4a7dd2040df1be121be8aa0f74dfc88665e9052"}
{"id": "train_21519", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "reproducibility", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "50272b8ed0efbaa9d47eb27a1a0404ccf4dd1f90"}
{"id": "train_21520", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "auditability", "security_gates", "ci_integration"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "83cdd776448499657509bc3c9fe6f560c62bd2be"}
{"id": "train_21521", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "security_gates", "documentation"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bea4c214cf1a535cf8e3bfb63d9638eeb8a485dd"}
{"id": "train_21522", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "reproducibility", "governance", "documentation"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "45a73414af96f96273de1a3162a5fa1e8fde911e"}
{"id": "train_21523", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "tests_are_truth", "auditability", "security_gates"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5c550ba6e26c1ae27c14fbb2d41c76d984851aff"}
{"id": "train_21524", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "documentation", "auditability", "reproducibility"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "02fd9970f134085c77a90ae8f2cdf1b40efa43be"}
{"id": "train_21525", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "tests_are_truth", "reproducibility", "documentation"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "76266b157514fbc4baf370245b15caaa7b9300e4"}
{"id": "train_21526", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "evaluation_metrics", "governance", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e939a7bb0b52f20dc5e07093495a493a7c480cca"}
{"id": "train_21527", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "governance", "ci_integration", "tooling"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c6d346493e9918d4a390bbfc819fa57633895b2d"}
{"id": "train_21528", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "security_gates", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "19cb97db7bb8a57bee9f9bd80e266ba0971c5ec4"}
{"id": "train_21529", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["governance", "cost_latency_tradeoffs", "security_gates", "evaluation_metrics"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6f4d4aa18e44f85a31930d2f3ad5cd75e0dcd604"}
{"id": "train_21530", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["auditability", "governance", "evaluation_metrics", "reproducibility"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dcafe30bd750d99da1a46d77e256fc42e2186842"}
{"id": "train_21531", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "auditability", "repo_scale_reasoning", "governance"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f68d78668aef1e9fd7d633f398b65a89f612e8d2"}
{"id": "train_21532", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "repo_scale_reasoning", "evaluation_metrics", "governance"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "116db2857333e87f8796849a13b9273cf208b092"}
{"id": "train_21533", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "tooling", "governance"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "01935252c47e813c99458b96a97698bc2b95debf"}
{"id": "train_21534", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "repo_scale_reasoning", "security_gates"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "399f81a490de539d95e49e5ff9830070b0045dde"}
{"id": "train_21535", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "tooling", "auditability", "reproducibility"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fb871a716f5b5c43978c48941244f4fa880b2351"}
{"id": "train_21536", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "documentation", "governance", "auditability"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f0809b3c4d2ddc3d9e3a19bd962fbf4796d04f19"}
{"id": "train_21537", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "tests_are_truth", "tooling", "reproducibility"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5f9a1436f57856a61f06cc52bf88d4f2599e0afa"}
{"id": "train_21538", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "ci_integration", "evaluation_metrics", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "65dddcded88f699e01ddf12aa6d1157ba5c110a1"}
{"id": "train_21539", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "documentation", "reproducibility", "tooling"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6402ae11041a37acf35b648ca94b56facc6ea042"}
{"id": "train_21540", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "tooling", "auditability", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f75796afbd69ad839a014586c13401988eb59122"}
{"id": "train_21541", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "documentation", "tests_are_truth", "tooling"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d52c1539bec2d113d01cfe13554fa23a5c224c9c"}
{"id": "train_21542", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1f8eaf8a31f9d13c643c31b27152c560dae3297d"}
{"id": "train_21543", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "cost_latency_tradeoffs", "tests_are_truth", "documentation"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "258b40472779a0131afc54e35b2ff7a2ee8f2ef0"}
{"id": "train_21544", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "ci_integration", "tooling"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "363aded3bbef925fbf26645b535fda72d777e960"}
{"id": "train_21545", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "repo_scale_reasoning", "reproducibility", "security_gates"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8ff5b3effb0a1a3e2ab3d79cd466f148978aa43f"}
{"id": "train_21546", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "documentation", "auditability", "tooling"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a06244f23fc693d1f724f1aecb8136f2e29b7f33"}
{"id": "train_21547", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "tooling", "reproducibility", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c900496b47f6412c51e4dc615a32497a9c6084dd"}
{"id": "train_21548", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "cost_latency_tradeoffs", "evaluation_metrics", "governance"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7b6155471c5a3ae3805fe7f26ae1ba183657bacb"}
{"id": "train_21549", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "tooling", "cost_latency_tradeoffs", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1aab9a3ff0e08e0f05fa4b4ff989fc674f3cbc6c"}
{"id": "train_21550", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "governance", "ci_integration"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "455589f1b643d013a1952f1b0b8376e812f1a84d"}
{"id": "train_21551", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fac9e864180915821743acd52cb2d57b96732b48"}
{"id": "train_21552", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "cost_latency_tradeoffs", "auditability", "governance"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "968a37c35c5ca0a1f790a2873fa92100b4f68538"}
{"id": "train_21553", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "governance", "auditability"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "df6fe9b8c9c4d06e2f5bbd3a95090889182013ef"}
{"id": "train_21554", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["governance", "documentation", "ci_integration", "auditability"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d4c493f93a3dfcf26928835f77523507c09c5591"}
{"id": "train_21555", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "security_gates", "tooling", "documentation"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1dd8a2ad1442328ee862c63cf0698eddb396c05a"}
{"id": "train_21556", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "cost_latency_tradeoffs", "reproducibility", "governance"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5b3c086d8b9b626baf1c8edd90bb9b09af041aaf"}
{"id": "train_21557", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "reproducibility", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "707b9ddc6896311cdfc29ba04e5348d46cbc5258"}
{"id": "train_21558", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "tooling", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "55e7003b115a0acf58dec68f75d1d67a929894e9"}
{"id": "train_21559", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "ci_integration", "reproducibility", "security_gates"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ff40f5e64d74c5302cd04f2768e7a1dcc21961c8"}
{"id": "train_21560", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "tooling", "governance", "ci_integration"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "059108832d5b5f5ccf3c09d247e51750be3512c6"}
{"id": "train_21561", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "tooling", "documentation"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d5e38ff3c2f9a90e7669d3458926f94f5e6a51f2"}
{"id": "train_21562", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "tests_are_truth", "security_gates", "ci_integration"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6f8961f208b2ef6663ed95e3b6c9ecd3dff665c2"}
{"id": "train_21563", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "repo_scale_reasoning", "evaluation_metrics", "tests_are_truth"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "47d4e0d3533eded66ac4473bcc01e8468b67917a"}
{"id": "train_21564", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["governance", "auditability", "tests_are_truth", "reproducibility"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "614d2a08e0c79ca6f3cc55db922e6a9381724e8c"}
{"id": "train_21565", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "tests_are_truth", "security_gates"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "beb41ec901300b8206653ab896d993c032b60c1d"}
{"id": "train_21566", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "security_gates", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d3fb3cc707d1d4b46b1f56c34ad5256807e8d449"}
{"id": "train_21567", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "ci_integration", "tooling", "governance"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "25d96d2f2057343ba197dab49e9f3588946adc61"}
{"id": "train_21568", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "auditability", "ci_integration", "governance"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a7d798087b4e91ce8b46c69bfbd161c6c5d2a477"}
{"id": "train_21569", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "auditability", "governance", "ci_integration"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3e373423bec2cf145b2617c7e3b79976b618927e"}
{"id": "train_21570", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "documentation", "tooling", "auditability"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2e129565ce285d33588c88d8f3342dca847173d4"}
{"id": "train_21571", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "cost_latency_tradeoffs", "documentation", "evaluation_metrics"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e2378f28ca40770f03ccb3a26729e39b0330816d"}
{"id": "train_21572", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "governance", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8346f1333705903b9875645d96fd82445a6426a4"}
{"id": "train_21573", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "governance", "documentation"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5d1a099401404fa2228fa99fd43f0496ccd92af1"}
{"id": "train_21574", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "ci_integration", "documentation"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0112017e6ae9b67a16cce2056db9effb99ae28a1"}
{"id": "train_21575", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "evaluation_metrics", "auditability", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4510cefbcded174946a44533b33e21635eb275d3"}
{"id": "train_21576", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "documentation", "ci_integration", "reproducibility"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f02c3111c3afc4add427a5145384b67976072586"}
{"id": "train_21577", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "governance", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "434fe5cd6392493f0bc2f648500dea8f0bb50120"}
{"id": "train_21578", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "cost_latency_tradeoffs", "governance", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a4d736ddb62bc7008988eb14c7cd107341382994"}
{"id": "train_21579", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "documentation", "tooling", "tests_are_truth"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "105855ea1a857bb2f9985be87c02899539bfd614"}
{"id": "train_21580", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "tests_are_truth", "reproducibility", "tooling"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ae1f700db0252932544f283350d3f29d837e3928"}
{"id": "train_21581", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "tooling", "reproducibility", "auditability"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7f71b5765a79eb532b9d0524c8566f209c143a79"}
{"id": "train_21582", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["governance", "reproducibility", "ci_integration", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "221d802af38eb1cd86fce9f34f66275c22c32a80"}
{"id": "train_21583", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "governance", "documentation", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5bf8b25a531877fda8e28c3bac4c4d9e0e26ffa5"}
{"id": "train_21584", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "ci_integration", "evaluation_metrics", "governance"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e3016e95a35681006fe5ed7e8b10f149c234dd0b"}
{"id": "train_21585", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "tests_are_truth", "ci_integration", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e5d94c5bd9a8f98f5b7f06698052cad6450ea40a"}
{"id": "train_21586", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "tests_are_truth", "auditability", "tooling"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "84603585ce645f45cafc952f9bacaaa31c2e7d4e"}
{"id": "train_21587", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "documentation", "security_gates", "governance"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f1b9b22ce5ead9fba3bbe763ba260621526d0797"}
{"id": "train_21588", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "governance", "security_gates", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ba3bb580e49a41bb5df246fbd0ba1041dba4952d"}
{"id": "train_21589", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["governance", "reproducibility", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b4057df8db491dbef69429bd733efad19e3db0c6"}
{"id": "train_21590", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "reproducibility", "governance", "ci_integration"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6cef8b51bc01e85e24c08ed89871c164a5540578"}
{"id": "train_21591", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "auditability", "governance", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f6ca956c8602f08540e4e26abfbc3e429431007c"}
{"id": "train_21592", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "auditability", "governance"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e676ee877b05e822ad58bfc87a9bbf08427f0b2b"}
{"id": "train_21593", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "auditability", "cost_latency_tradeoffs", "security_gates"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c6198e265c405ae699db6efdeb2deb05ca5c3478"}
{"id": "train_21594", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "tests_are_truth", "cost_latency_tradeoffs", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c541cfad2747a1b52c2a449e5a24bc8153dc4eb0"}
{"id": "train_21595", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "cost_latency_tradeoffs", "governance", "evaluation_metrics"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "73be2c783bcc0d25bc73091639003dd29d7f4925"}
{"id": "train_21596", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "governance", "tooling", "documentation"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9d64e93cd58e6764bc9e7258f35e5c3c4119eff6"}
{"id": "train_21597", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "reproducibility", "documentation", "security_gates"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6d568005ec938b683e33c517f3b86c96c3ef5cd8"}
{"id": "train_21598", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "reproducibility", "evaluation_metrics", "governance"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4de77d2cb0c331ece21c341002b929be2b24f852"}
{"id": "train_21599", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "repo_scale_reasoning", "tests_are_truth", "ci_integration"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "597aef5cc136db30b3712ac5edb4705cea49ae78"}
{"id": "train_21600", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "auditability", "tests_are_truth", "reproducibility"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "aece5cd83cbe25c14c74fd7995ee22522f2fad06"}
{"id": "train_21601", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "repo_scale_reasoning", "governance"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2910c0cd8e9ce0a987b7426743a9bee0048e5308"}
{"id": "train_21602", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "security_gates", "tooling", "governance"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "87f8335fada390bde2ea0cdc9fb97b1beb595103"}
{"id": "train_21603", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "reproducibility", "auditability"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "49b8c1397df791513a95737e7752215282147a03"}
{"id": "train_21604", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "evaluation_metrics", "documentation", "auditability"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f4baf668c8f899d3518f53845fcd1fa885a82b6d"}
{"id": "train_21605", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "ci_integration", "tooling"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "62f34e898772ae3d6b78f21edd3f0d7f0cee888a"}
{"id": "train_21606", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["documentation", "tests_are_truth", "auditability", "tooling"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3312ab162ad54ea0323c2a071cd1afa20b0157e1"}
{"id": "train_21607", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "tooling", "tests_are_truth", "reproducibility"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cc36297859c1bc8826152ce9bb04f0db0bc86537"}
{"id": "train_21608", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "documentation", "evaluation_metrics", "ci_integration"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2869875a5511c5c9b0e7dea837f6f1c979dc6783"}
{"id": "train_21609", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "tooling", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3199fffb3987204494667efa7be37461753df528"}
{"id": "train_21610", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "evaluation_metrics", "tooling", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "34981987cb89fb5748fd7c67c2451e5cb2a53391"}
{"id": "train_21611", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "auditability", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4e00b3c208bbeb2af75ee8b6861b79e7e1984889"}
{"id": "train_21612", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "reproducibility", "evaluation_metrics", "auditability"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0b02f779d3922a9d384cf7c8d6b89983a90ee316"}
{"id": "train_21613", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "ci_integration", "evaluation_metrics", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d0d12e791c26d5a2dafe76e7096f877243beaf6f"}
{"id": "train_21614", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "documentation", "cost_latency_tradeoffs", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "be293c127632777fd7726529c98d6d108a453499"}
{"id": "train_21615", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "auditability", "repo_scale_reasoning", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "742bc4b3237368787e7c126d55ac9268aeacfa60"}
{"id": "train_21616", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "governance", "auditability", "documentation"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0fafda2aa06bd77d54a1554139b5828c9296b02f"}
{"id": "train_21617", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["governance", "reproducibility", "auditability", "ci_integration"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2878dd91d5796bc9b1a17ef72e6b5e963ee21754"}
{"id": "train_21618", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "repo_scale_reasoning", "governance"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e86658845f996422fc6e2ad9602b62de2f8284a8"}
{"id": "train_21619", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "security_gates", "repo_scale_reasoning", "reproducibility"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "928bac86082b34cbfd969ee0b9e447accbec4732"}
{"id": "train_21620", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "security_gates", "tooling", "ci_integration"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fe3e926ba5abaef629d0bd825aad9d08dd81470c"}
{"id": "train_21621", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "ci_integration", "tests_are_truth", "auditability"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c59e0b93d367bd52f8bcd87dd56d5b29c08a0bb0"}
{"id": "train_21622", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "tooling", "ci_integration", "governance"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7917c61882acf73d23c9f15597d390d81a559b1f"}
{"id": "train_21623", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "security_gates", "repo_scale_reasoning", "evaluation_metrics"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b3b8450566ee90251006d274fde99839b0e07b3b"}
{"id": "train_21624", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "repo_scale_reasoning", "reproducibility", "ci_integration"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e30027a9304757d4ee801ffb4d5bab3e8fdd4b96"}
{"id": "train_21625", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "documentation", "tooling", "security_gates"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6bcde5cbc8021aa664b9e71b5bec113bd611f837"}
{"id": "train_21626", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "security_gates", "evaluation_metrics", "reproducibility"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "85380657f7568c1a33d2778b05ebc0fe3f0b4efd"}
{"id": "train_21627", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "tests_are_truth", "reproducibility", "documentation"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "86fdd2522d2aa69b49a16538ac36419d7524ff6c"}
{"id": "train_21628", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "tooling", "governance", "security_gates"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cc4cb2e4475f2adae71f66617de91d87c58fe8c7"}
{"id": "train_21629", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "repo_scale_reasoning", "evaluation_metrics", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "be3971e982bcfb8bdb66fc058df59c46f5c37f85"}
{"id": "train_21630", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["governance", "security_gates", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bc9b0114ceb47542ca127cedcd20237d45c56484"}
{"id": "train_21631", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "auditability", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "74a92bbe1c3fe994807ec0572076fb15eb875440"}
{"id": "train_21632", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "evaluation_metrics", "tooling"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b13a800a63d2609fbe52c1a967a3781569ec2e4d"}
{"id": "train_21633", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "tooling", "reproducibility", "evaluation_metrics"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "da3059de1e90dbf9906307dea1de6d7d66f4ebb4"}
{"id": "train_21634", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "evaluation_metrics", "governance"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "56d68f00b629e6d8087b9968114f190060e2e437"}
{"id": "train_21635", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "tooling", "governance", "tests_are_truth"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2456b66f7bfc4ab5ede829d8675134149fb5fa1e"}
{"id": "train_21636", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "repo_scale_reasoning", "governance", "documentation"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "12188c90dff3e009d85b9745cb863729fb899342"}
{"id": "train_21637", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "tooling", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "084d50dc4a82a36cc7311af1fdd9e8ca7ce9b3ad"}
{"id": "train_21638", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "governance", "ci_integration", "security_gates"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f0d492b70ee432779394eb8f3be6c1cf64a15c60"}
{"id": "train_21639", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "repo_scale_reasoning", "evaluation_metrics", "tooling"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5a053dbed3b01194d4e200092d4c27df76cfd4ff"}
{"id": "train_21640", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "tooling", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4a5121faa79310d7b5a58cc758560e493f547227"}
{"id": "train_21641", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "tests_are_truth", "ci_integration", "governance"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c5b87c1382faf0a4c01f1a8a4a4342b22fb34bab"}
{"id": "train_21642", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "governance", "repo_scale_reasoning", "documentation"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7ae359be7b8fa6c9ad5f25e9d18a1e3fd308339d"}
{"id": "train_21643", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["auditability", "security_gates", "evaluation_metrics", "repo_scale_reasoning"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7a307522e883f0711b0e1408c95738a55266c388"}
{"id": "train_21644", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "tooling", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3b1ea30b5874bb5b6e68d1c2117f3325451d9e3d"}
{"id": "train_21645", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "evaluation_metrics", "security_gates", "documentation"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4568a6a944678b12a5cf11e1d270ab29b1edeaaa"}
{"id": "train_21646", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "auditability", "tests_are_truth", "reproducibility"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "65d6c1a681f3da4e6dcccf4ab7f8a04c7acc97b2"}
{"id": "train_21647", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "documentation", "ci_integration", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3da8fb281cbc6a34d2458f06c76633b6393d17ab"}
{"id": "train_21648", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "tooling", "security_gates", "evaluation_metrics"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0cd6c5dac86e3e5b7d05cb650a9d490920cd830d"}
{"id": "train_21649", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "auditability", "repo_scale_reasoning", "reproducibility"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bedb944450130fbe707716004079938b223ffe21"}
{"id": "train_21650", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "governance", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f371044665103d7851bfbce6581fe856bde29fc7"}
{"id": "train_21651", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "reproducibility", "security_gates", "governance"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bc5054e4612c42f0d4762273ce20477edce6a0a5"}
{"id": "train_21652", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "reproducibility", "auditability"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e6088d4fa84b335832be52c9287e494184f3cc26"}
{"id": "train_21653", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "repo_scale_reasoning", "governance"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "aaad428bd4f9ac772a607243114c7377a2eac8ab"}
{"id": "train_21654", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "auditability", "governance", "tests_are_truth"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "91c2150330209edc3489df8a957cdfdd82826e77"}
{"id": "train_21655", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "tooling", "evaluation_metrics", "documentation"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0ba9e42bb469842b6ef61b2979c81fa1b5c7f3d8"}
{"id": "train_21656", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "documentation", "governance"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0fa2ce7c1e6552cdbf2e793d916dc12e66e44518"}
{"id": "train_21657", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "reproducibility", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "084d50dc4a82a36cc7311af1fdd9e8ca7ce9b3ad"}
{"id": "train_21658", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "ci_integration", "repo_scale_reasoning", "tooling"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "10bc942b01ab15e0c0af2ad8f3bf0f8a8fa84151"}
{"id": "train_21659", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "security_gates", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "77915ff6fc2b9b4063ce701a6f8d92c4b2d0edb2"}
{"id": "train_21660", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "security_gates", "governance", "documentation"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6744f99415ffbc33be867e82619536ec3e62bc51"}
{"id": "train_21661", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "evaluation_metrics", "repo_scale_reasoning", "auditability"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ed82ebf29338d6f95aeffe5510607df9a022b874"}
{"id": "train_21662", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["governance", "ci_integration", "tests_are_truth", "reproducibility"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a64601eae068bcb538b958d2309e501dc4f3c138"}
{"id": "train_21663", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "auditability", "security_gates", "governance"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "42ff26ae17345ce63726ac4469edefcc9b9af93d"}
{"id": "train_21664", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "security_gates", "repo_scale_reasoning", "documentation"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1eb59525d087f4a83ef1cd6a1d4d1a49ec0d631b"}
{"id": "train_21665", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "repo_scale_reasoning", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "81808e553142096e1e5584dcfe77ba318ddb414b"}
{"id": "train_21666", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "reproducibility", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "286f2fe3e8b8a37b3ea881d525a2cf279c675a3f"}
{"id": "train_21667", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "tests_are_truth", "auditability"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b5c799b64bb68c0092bbc2c0c1b79e958b2732cf"}
{"id": "train_21668", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "auditability", "governance", "tooling"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a5c8b637cb8b3ac5633542199eb0043edc6e9d6c"}
{"id": "train_21669", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "security_gates", "tests_are_truth", "auditability"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ea851455f6e0c024fb3ab09b6882d22517403c20"}
{"id": "train_21670", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "repo_scale_reasoning", "security_gates", "reproducibility"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ca62f811b6b069b988fc3af1819c26034957bd31"}
{"id": "train_21671", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "repo_scale_reasoning", "security_gates", "ci_integration"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9500c717c13236f2837f2cadaf17cbd26b01359d"}
{"id": "train_21672", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "repo_scale_reasoning", "evaluation_metrics", "reproducibility"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ca7a4a3c04b26b713e6569c3335da9ad3cf8404e"}
{"id": "train_21673", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["governance", "security_gates", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e248886462738d13c07f282c470da9389e8973ee"}
{"id": "train_21674", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "documentation", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "918f2ecf88bace9ecc9a904aa73afc6fc0e3c2da"}
{"id": "train_21675", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "tooling", "reproducibility"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5038067eaa7d3e25f1d54fd08642dc6117f97f4c"}
{"id": "train_21676", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "tooling", "tests_are_truth", "reproducibility"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7a9aa53c7fc57fcaae65bbb8a5fabeaa2c4fb331"}
{"id": "train_21677", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "auditability", "tests_are_truth", "ci_integration"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0bb1cb3b4d3af417ab7d74585ce04dc1b0f38d19"}
{"id": "train_21678", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "reproducibility", "tests_are_truth", "governance"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6aec82bf6f2abd0ae99fb41d6db090f2a1a28f8b"}
{"id": "train_21679", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "tooling", "security_gates", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7a02e06304f603a9f7b0a5330ac98ae399651d1b"}
{"id": "train_21680", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "ci_integration", "tooling", "repo_scale_reasoning"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "610541442ffd2b1f1c3f2e008d3d1749ab5af6c7"}
{"id": "train_21681", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "reproducibility", "tooling", "ci_integration"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "97d0020dbbf187e0a317782b411d91b263998092"}
{"id": "train_21682", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "tests_are_truth", "ci_integration", "reproducibility"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9692da5ea434dfbe4ad7ebfcaf62589345dc086d"}
{"id": "train_21683", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "auditability", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c68a9e5bfc00f2cfd4b383489d7ac3c12feaa4a2"}
{"id": "train_21684", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "tooling", "evaluation_metrics", "reproducibility"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "75c812bb4fc6222d419987550d1fb07711ddbe3a"}
{"id": "train_21685", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "reproducibility", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cfef453119f831d60eb685c6e4f14104b0e48286"}
{"id": "train_21686", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "tests_are_truth", "tooling"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "71fce73349f3330ae2709f7d4d1f444761ef7f8b"}
{"id": "train_21687", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "governance", "auditability", "ci_integration"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7cca1b52424dcf4d854b2b386ff8851864c3014c"}
{"id": "train_21688", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "repo_scale_reasoning", "tests_are_truth", "tooling"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3707eaf7d2814f3d8599dac4512f88c8ea00c4ec"}
{"id": "train_21689", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "documentation", "governance", "tests_are_truth"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a1de2a123afe7c222d795ebaed877445cc3f8853"}
{"id": "train_21690", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["documentation", "reproducibility", "governance", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "957048ba966c37be888d2251ad2e47c0bc804257"}
{"id": "train_21691", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "reproducibility", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fa949f6799d94773ff16ec92ccbc3f1b47ceeb25"}
{"id": "train_21692", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "documentation", "governance", "auditability"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "db1ebd501ab9892dd1c98785de2f30629350312c"}
{"id": "train_21693", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "auditability", "governance", "ci_integration"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bf574033fdd9618ec2b82cee5cb2306acca2d84f"}
{"id": "train_21694", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "auditability", "security_gates", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "be0f8c5c001f27ceb9a314f44b7ce8c4b03b9817"}
{"id": "train_21695", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "documentation", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8f12d739a41ef6d962a203cffd17417ce1979826"}
{"id": "train_21696", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "ci_integration", "security_gates", "auditability"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6344926da694befb5d226954dbb15ae4b94ba947"}
{"id": "train_21697", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "repo_scale_reasoning", "security_gates", "ci_integration"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5d0b0677e872bcb8e2371cf04b708240641e9dad"}
{"id": "train_21698", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "cost_latency_tradeoffs", "ci_integration", "tests_are_truth"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5f99d9680da1414912bfd8618dbf4ebd3c671c23"}
{"id": "train_21699", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "evaluation_metrics", "repo_scale_reasoning", "security_gates"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ff36a36ebc7fcf51cdbf35fee3e2a9ca72664948"}
{"id": "train_21700", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["governance", "reproducibility", "auditability", "ci_integration"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4aef6afcd8395a42d22e72f46af1aac04e88bc0f"}
{"id": "train_21701", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["documentation", "cost_latency_tradeoffs", "auditability", "security_gates"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c184a19a6ca0171932f3066fe444ce116342ea18"}
{"id": "train_21702", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "tooling", "cost_latency_tradeoffs", "governance"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b71ac3dc51c8ddeba5d275bb175daf16492047bc"}
{"id": "train_21703", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "tooling", "repo_scale_reasoning", "auditability"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "772014e51e61279602e4156527fe9ce029b23b39"}
{"id": "train_21704", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "security_gates", "reproducibility", "ci_integration"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "97f95447c275faeb1c4b6c422d144b489904160c"}
{"id": "train_21705", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "governance", "evaluation_metrics", "auditability"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "50d83363fb252a9b0f39fb7ecb04b01b61529efa"}
{"id": "train_21706", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "governance", "tooling"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d3c975f17367d4d85a998525a0441e1d213eed06"}
{"id": "train_21707", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "governance", "documentation", "reproducibility"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a7b05aed4c763fcc9b3a632609e4d73bf73a70cb"}
{"id": "train_21708", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "56213a35dbfacb2127fd6320e64ebc80ee22912d"}
{"id": "train_21709", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "evaluation_metrics", "tooling", "documentation"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "16268dd8d094ead5ba5cbb81543f2c82db3c6d01"}
{"id": "train_21710", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "evaluation_metrics", "auditability", "ci_integration"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "42ca21e67b76fb6d442af95ce756ad1f33bed0c8"}
{"id": "train_21711", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "auditability", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f0b95fdbb49c8d5fa3211b5dcfc5056a7a243a36"}
{"id": "train_21712", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "evaluation_metrics", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e08d1a49b5ded90c8ee985e1e57fd2dae08f389e"}
{"id": "train_21713", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "governance", "reproducibility", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5a8711c515613be5095d04e2a5687e98b0b3dfa8"}
{"id": "train_21714", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "ci_integration", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "53b256c53d05e666131c3b93e1fb857a05136598"}
{"id": "train_21715", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "documentation", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1f6881c253590cf01c3a73a165ef0b56be72f6be"}
{"id": "train_21716", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "governance", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3f02d062cc87e6778736de10265fc82ef2a4b5d8"}
{"id": "train_21717", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["governance", "security_gates", "tests_are_truth", "tooling"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cdc173b23a1b83f706773cfe3a9dfe82d9684ec5"}
{"id": "train_21718", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["auditability", "reproducibility", "governance", "tests_are_truth"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dfc215b7bffe58408ea1dd9af62c1799461364ec"}
{"id": "train_21719", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "governance", "documentation", "tooling"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "10a9015957c2f3d8ffa71f7806987016c2e4f482"}
{"id": "train_21720", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "cost_latency_tradeoffs", "ci_integration", "security_gates"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0185ca9129cd5cb8fa9aec450bf9b66a3ad849e6"}
{"id": "train_21721", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "documentation", "reproducibility", "security_gates"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e60a641ca130837a8025326e7acafa36bfe2bb4b"}
{"id": "train_21722", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "reproducibility", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ae40a7edfe4a8c365319831f8f5ef863d6875ac2"}
{"id": "train_21723", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "documentation", "tests_are_truth", "governance"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "542bd5144fbfb125b86d3123a4951b3715b6ab79"}
{"id": "train_21724", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "security_gates", "evaluation_metrics", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9116f694bebc08f82d6652c169ddc4dc68a8e2c5"}
{"id": "train_21725", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "documentation", "security_gates", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ed217ae749c39263a54cc08f7568601f099b3a5e"}
{"id": "train_21726", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "auditability", "ci_integration", "reproducibility"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "24e44acb884ea7669af23150b5064e3218b08d5b"}
{"id": "train_21727", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "ci_integration", "governance", "reproducibility"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "81b1d78442360d804d7c0a7fd5acc6ba2be13bc1"}
{"id": "train_21728", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "documentation", "repo_scale_reasoning", "governance"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b9903ea4ab1b07adb740df25bcd1deeec7e6c28e"}
{"id": "train_21729", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "governance", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9e49a27f8fa441d10b7459b195fef99225e692d1"}
{"id": "train_21730", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "cost_latency_tradeoffs", "tooling", "auditability"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "915640a0bfe795fe4da66a8359f3dcb2888ccbe0"}
{"id": "train_21731", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "tooling", "repo_scale_reasoning", "documentation"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b9aca4dbbd26e307b1d0aff8ec3294505bb74354"}
{"id": "train_21732", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "tooling", "governance", "ci_integration"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "42fe14bb145bbde31a33af4b3cbbed84c26a05db"}
{"id": "train_21733", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "evaluation_metrics", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bc7a8454d4bfe4080131ac1a52ea78e9c0bf3f4a"}
{"id": "train_21734", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "repo_scale_reasoning", "auditability", "security_gates"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "363b45981050e103e69d912f4959a8979ace0b7a"}
{"id": "train_21735", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "evaluation_metrics", "security_gates"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "38ab4f9d6838203c5de30a08fe8b8afb40c6f744"}
{"id": "train_21736", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "ci_integration", "repo_scale_reasoning", "evaluation_metrics"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "78e753edc73d2c833fd72b6d38191b754d6294ff"}
{"id": "train_21737", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "tests_are_truth", "security_gates", "auditability"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "38dd67098826706fe61defb6e0d388d7c9014198"}
{"id": "train_21738", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "tooling", "auditability", "reproducibility"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f75bd2c1e731e7bf9349677c9301451c5e60aa25"}
{"id": "train_21739", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "security_gates", "cost_latency_tradeoffs", "governance"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2f4d26d1eb57c15c768cb3761e8565b97d0293e9"}
{"id": "train_21740", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "auditability", "tests_are_truth", "governance"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7bd2de388ab7a2d223249f05e2033956a59fa048"}
{"id": "train_21741", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "repo_scale_reasoning", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "548a8b0ee02b5f05a83bf271a2b4d19137e39970"}
{"id": "train_21742", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "security_gates", "documentation", "auditability"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a6ec48c4c10b8be47d6165e45830487e607dce36"}
{"id": "train_21743", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "tooling", "reproducibility", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f74729b154621a0b53c94b6cc8cec3166aab3458"}
{"id": "train_21744", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "security_gates", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0d76d0e3bf459fe370808979489c31ea566255be"}
{"id": "train_21745", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "ci_integration", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "539e7df81984e833ac62d7c25e3c5032873ef730"}
{"id": "train_21746", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "reproducibility", "auditability", "security_gates"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b6a20c482766e1fb7c9abe0f85fd1ca0455a8437"}
{"id": "train_21747", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "tooling", "reproducibility", "tests_are_truth"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6bb1c5531f25a0a130dccdcad4c45ca21d09902d"}
{"id": "train_21748", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "reproducibility", "tests_are_truth"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "014577cdcbb46e15e55209c9aaf23682d2decc27"}
{"id": "train_21749", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "documentation", "auditability", "reproducibility"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "af9241dbb5b48923c2ad5ef80ed48f8e6830cafd"}
{"id": "train_21750", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "cost_latency_tradeoffs", "evaluation_metrics", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e4b351df7b9c3636229ec5cc574a8e7c759bc175"}
{"id": "train_21751", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "ci_integration", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1dbd8e046497381832ee6a859de3abe81a1f6c88"}
{"id": "train_21752", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "documentation", "tests_are_truth"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4ced659436272de600aa54672950613a661804ff"}
{"id": "train_21753", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "repo_scale_reasoning", "documentation", "ci_integration"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "16145593afece0d7472e0b015d0bd324b424ca85"}
{"id": "train_21754", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "governance", "tooling", "auditability"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3c0fc400c01e69e9bfc913a1d8304ab8d6dff1ed"}
{"id": "train_21755", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "tests_are_truth", "governance"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f593b09fe5d2260f64582f80d96aff73d24683d1"}
{"id": "train_21756", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "tests_are_truth", "security_gates", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "587688181aa5d268164a06e107d6b10d205bcf89"}
{"id": "train_21757", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "tests_are_truth", "ci_integration"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0f6a5c6c8044c08a70f0a49ad1ca6fbe351e76b0"}
{"id": "train_21758", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "governance", "ci_integration", "tooling"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "469a14d3a59d0f5d5202893c7a299b87aad54df7"}
{"id": "train_21759", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "auditability", "reproducibility"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8cea2f0ab13e461fc463ffaa8db4e025513ec92c"}
{"id": "train_21760", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "auditability", "tests_are_truth", "documentation"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "416bb8b7c1d0bc0bc7eca5bbc3adf3b6583e5005"}
{"id": "train_21761", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["governance", "auditability", "tests_are_truth", "documentation"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6ffc1abae483e05867b6414ccc39ce0c2f9afffb"}
{"id": "train_21762", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "tooling", "governance", "tests_are_truth"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "17d9a8df3a4638283a55a9c20c39095a1a557277"}
{"id": "train_21763", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "repo_scale_reasoning", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7905d90d3d9a3b4015730555e700956044c71302"}
{"id": "train_21764", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "tests_are_truth", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b1c8d7009406d784fdb651cdc16f7aa59024e22a"}
{"id": "train_21765", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "documentation", "tooling", "security_gates"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f4a2b2aabd9ecdea96b4992a8af5dc8411338891"}
{"id": "train_21766", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "security_gates", "evaluation_metrics", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7823147a4a63257df8da6a46aa028bc1132272b7"}
{"id": "train_21767", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "evaluation_metrics", "reproducibility"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cd8d3c195ee424e06fe2bf53a0b611d0a4ca6d9d"}
{"id": "train_21768", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "reproducibility", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5a1d6e4f51d08f1d00d40857d3971a736f89b93a"}
{"id": "train_21769", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "tests_are_truth", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d25d7594badac421718820a2f03b2d6d5050f4d2"}
{"id": "train_21770", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "ci_integration", "documentation", "tests_are_truth"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "62162af32264eb20e045cd71aa7f34c1caa8b039"}
{"id": "train_21771", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "security_gates", "auditability"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e69eb85ce3eb58e73dcd3247a977e009cda323b7"}
{"id": "train_21772", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "documentation", "repo_scale_reasoning", "reproducibility"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "50ad995ae6acaf0657dc41118267226dddaa4cfa"}
{"id": "train_21773", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "repo_scale_reasoning", "tests_are_truth", "documentation"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "086bb3fff93504c68f9cc140c7416adeb01562de"}
{"id": "train_21774", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "repo_scale_reasoning", "tooling", "ci_integration"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ef9db386931de591738c1d6959bdbd6f01502ad3"}
{"id": "train_21775", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["auditability", "repo_scale_reasoning", "evaluation_metrics", "tooling"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7d91d85cca2ae59159967dccb0053c27ee7a133f"}
{"id": "train_21776", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "tooling", "reproducibility", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "89ef97748a4eac217bd3d775612edabc4cc04c05"}
{"id": "train_21777", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "governance", "documentation", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "12fe1d0a85373b8fc52d2aa1ac44736924633daa"}
{"id": "train_21778", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "tooling", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d3c459fd81d885280826bbcca5587fc491269c18"}
{"id": "train_21779", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "auditability", "governance", "evaluation_metrics"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "73f864828bad62e2fad41daf1d52ca0d7988c017"}
{"id": "train_21780", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0350711c692d7c946eaf453f9a58001db6d4bc00"}
{"id": "train_21781", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "repo_scale_reasoning", "tests_are_truth", "governance"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b3b9eccfc621920e1d88222bf47b1afa9788d250"}
{"id": "train_21782", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "tooling", "auditability", "security_gates"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6f780089eca27033f1496aca92ef30fe5cfdcff9"}
{"id": "train_21783", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "evaluation_metrics", "repo_scale_reasoning", "reproducibility"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ddd02f85e5bcd42bf2e3dea1fcf6e4a9c60cf44d"}
{"id": "train_21784", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "ci_integration", "reproducibility"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7f9e499609ea62d930ae2768d359546eaa77d711"}
{"id": "train_21785", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "tests_are_truth", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "86a2490d640a8074324ae36449b2a62e14f1331a"}
{"id": "train_21786", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "security_gates", "evaluation_metrics", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "98eae7fcbe1344ed036bfe31552da3a13e9753e9"}
{"id": "train_21787", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "ci_integration", "tooling", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fad51c712826ecc3aaba0510caf5374bd882ce21"}
{"id": "train_21788", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "repo_scale_reasoning", "security_gates"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5bf35161b06acf436ddd66e2e199c95350357002"}
{"id": "train_21789", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "security_gates", "reproducibility", "ci_integration"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cc68e4531542771d4890c22490f2651b459cb968"}
{"id": "train_21790", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "tests_are_truth", "governance", "security_gates"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f50af981a6a646f109214aab81bdbd4d37cf30aa"}
{"id": "train_21791", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "documentation", "tests_are_truth"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "804d4cb051d59705cd9a802898e22e8f13407ed9"}
{"id": "train_21792", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "tooling", "governance", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "95e0aaa3953feacd4aa9ae4c65abdbcc8f1d8204"}
{"id": "train_21793", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["governance", "documentation", "ci_integration", "tooling"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "642582b960231eef4b2ec382dc10edd8c8e59d72"}
{"id": "train_21794", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "auditability", "ci_integration", "security_gates"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0c3c296c0e568359aa3062ee46bf327e0fa1b759"}
{"id": "train_21795", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["governance", "security_gates", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6ffdc5ff26d73bcc35c4f21268bbc93b96db0eaf"}
{"id": "train_21796", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "tests_are_truth", "reproducibility", "documentation"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ca5ac0582c1aa918dad80fcbda56bbbf850069a9"}
{"id": "train_21797", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "auditability", "tooling", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a22806f2a78665064ae1d3fd3063f32b8b0d870a"}
{"id": "train_21798", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "evaluation_metrics", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "afdd3b82d923895fc27dd3f1f8385ede9ef6bd44"}
{"id": "train_21799", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "tooling", "repo_scale_reasoning", "security_gates"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7f88163b1b1896ae680c25d6b88d185697d6e045"}
{"id": "train_21800", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "governance", "tooling", "documentation"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b5b4ede8b08424002a7aad29eaf31295d9cbb16b"}
{"id": "train_21801", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "tooling", "auditability"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "50f449d6f9eb41588c6afeca595d36e98735a6ee"}
{"id": "train_21802", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "tests_are_truth", "security_gates"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "964b3332bab6e8d113c1c7b000a19b60d9848505"}
{"id": "train_21803", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "reproducibility", "documentation", "evaluation_metrics"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "540eba8f742735870cb183c23e34215535507d1a"}
{"id": "train_21804", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "documentation", "governance", "auditability"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bcbe9dfd2586fb477b9c8b1a79883dc57d586a84"}
{"id": "train_21805", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "documentation", "tooling", "governance"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "eedc604b6ec4162eb2169b2f44311b19f3a8c7ec"}
{"id": "train_21806", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "tooling", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dd6843281ee952a156ffe34cb9de205d66a5ed76"}
{"id": "train_21807", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "tooling", "documentation", "auditability"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "90b603fd0095212864ff94a9e86c876a71d2ba4c"}
{"id": "train_21808", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "documentation", "tooling", "ci_integration"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f51b7baf35966e9a2b2e69292597c1144b3bd9eb"}
{"id": "train_21809", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "governance", "reproducibility", "security_gates"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e842911fdc08e4521fcbbecfb3fbde820b9750e4"}
{"id": "train_21810", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "governance", "tests_are_truth", "documentation"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ff3ce01e215c73920fbe7488dff358a030bbed3d"}
{"id": "train_21811", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "governance", "evaluation_metrics", "tooling"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "136443340a458ec290aeb715b0b29aba2796dd64"}
{"id": "train_21812", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "repo_scale_reasoning", "governance"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3de555f230bc051c7b925b94d9c3ce8083e34fd4"}
{"id": "train_21813", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "reproducibility", "evaluation_metrics", "auditability"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0be6add95405c79cb3665a0bfd2681dc69080c2d"}
{"id": "train_21814", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "evaluation_metrics", "ci_integration", "governance"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9f3c890a5fe8efcb618f5598df631a193f256429"}
{"id": "train_21815", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "tooling", "evaluation_metrics"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "535ad9fd4a007bb711c1748289a09d2da9c5d85e"}
{"id": "train_21816", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "documentation", "governance"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1fce3fd40fb2271c83baab7a401775ad62ac43e3"}
{"id": "train_21817", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["auditability", "documentation", "tests_are_truth", "security_gates"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "18be958d7eb730540c7ba063d8be1e64be049afd"}
{"id": "train_21818", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["auditability", "governance", "reproducibility", "documentation"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "eb48a5a771e86f4e725b495f468fae4cfbfa1e7d"}
{"id": "train_21819", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "tooling", "tests_are_truth", "ci_integration"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "549db2b2d1c5ea6372ab52672d077a048ed9303a"}
{"id": "train_21820", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "reproducibility", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c0e6e9230391d670938bfb7b6b29bf45fd79f861"}
{"id": "train_21821", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "security_gates", "reproducibility"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0746a0b81e7f1882159cb5d7f3ef49a6937ccad2"}
{"id": "train_21822", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "documentation", "security_gates", "tooling"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "85faf0e358f745f27f142c7d2aaaf4cb03af7210"}
{"id": "train_21823", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "cost_latency_tradeoffs", "auditability", "tests_are_truth"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dfc215b7bffe58408ea1dd9af62c1799461364ec"}
{"id": "train_21824", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "documentation", "tests_are_truth"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7823b34a9d3b49f2120ba4ee7ad570b23f85f0cc"}
{"id": "train_21825", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "reproducibility", "evaluation_metrics", "tooling"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c15c1216978dfea0fc55f49a0aa7aa20b26c1943"}
{"id": "train_21826", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "auditability", "documentation", "repo_scale_reasoning"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d71cf8c61663bf6a507804a01299906b205ac1a6"}
{"id": "train_21827", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "documentation", "reproducibility", "security_gates"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f7cdb14a348e63ea72def56d9352a4dfe9039a83"}
{"id": "train_21828", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "governance", "tooling"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5c83b88da2c7c8428f53f58650bf0551b39ecc1b"}
{"id": "train_21829", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "governance", "tooling"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a075d48966a37c5243c7056793ef552592f9c097"}
{"id": "train_21830", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "tests_are_truth", "evaluation_metrics", "governance"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dfc0ddb1fa04aa97fca19c3677007f627b1cb04d"}
{"id": "train_21831", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["governance", "security_gates", "ci_integration", "documentation"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "94c23b67986b37d56e0537ec3a5f4a519f808fa0"}
{"id": "train_21832", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "documentation", "ci_integration"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bdbf4e7c36b32f6689ec7272a4a4794a16132f57"}
{"id": "train_21833", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "security_gates", "documentation", "reproducibility"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "115eed8757156b2a2161ced46c8a18fe247f4b5f"}
{"id": "train_21834", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "repo_scale_reasoning", "auditability", "reproducibility"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "78ebf1895ff2e7cdcf48fc91649fc946c3e3b6e1"}
{"id": "train_21835", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "tooling", "governance"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5a273409d7aa463dbe7c6a7baf6200288f5aff02"}
{"id": "train_21836", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "cost_latency_tradeoffs", "governance", "tests_are_truth"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bfaf866577e7e73d0962ae0d0e03ff70bd5c9ac7"}
{"id": "train_21837", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "evaluation_metrics", "auditability", "ci_integration"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0d57225d3f9fed2a586aace31f39872aa6a04ab0"}
{"id": "train_21838", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "governance", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5aeacd2bbe23905fc594352b04849c173faac668"}
{"id": "train_21839", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "tooling", "governance", "ci_integration"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "188271a32c3e11d0ab084abc5f24cc8eac8036ba"}
{"id": "train_21840", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "repo_scale_reasoning", "evaluation_metrics"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7b3dc66b1839e26d350b61cf36a79fa2b3a0d893"}
{"id": "train_21841", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "tests_are_truth", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6607b67e6964cc1e3980f764917cc7cd39868419"}
{"id": "train_21842", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "cost_latency_tradeoffs", "ci_integration", "evaluation_metrics"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e4e520b757b1e6b57d6fe55223e2cb7c7f345191"}
{"id": "train_21843", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["governance", "auditability", "evaluation_metrics", "tooling"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b8870f863d614d2b60258e713449cf6ffba909b7"}
{"id": "train_21844", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "governance", "ci_integration", "auditability"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "66fb9fd1c041d6f3cc229b738aa02309f5e0c2a3"}
{"id": "train_21845", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "governance", "tooling", "auditability"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "aa63c461c806e52eaf29b3d722731e08e0461b5b"}
{"id": "train_21846", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "security_gates", "evaluation_metrics", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1901a39a900a2dfc0fe04ec960eaba463d5c4367"}
{"id": "train_21847", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "governance", "security_gates", "tooling"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "37dde2c24e7736dbfeb93a8dc42ea00851b45526"}
{"id": "train_21848", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "reproducibility", "governance", "auditability"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "416ee131a96f5a54be9fcc10ef06ffb8eeb8b8b7"}
{"id": "train_21849", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "auditability", "tests_are_truth", "security_gates"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "747b02c225e44dd84f38ad090f7d8927025bc671"}
{"id": "train_21850", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "ci_integration", "reproducibility", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8810cecb0f8c25e50536c461134f424edd8cf7d0"}
{"id": "train_21851", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "documentation", "ci_integration", "repo_scale_reasoning"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "71dbf064ab7a0b8a13712365fc71168f1fed8432"}
{"id": "train_21852", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "tooling", "ci_integration", "security_gates"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "33b89348338c3fd76334ecb77f5f7680b3a40239"}
{"id": "train_21853", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "tooling", "repo_scale_reasoning", "security_gates"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7e1d2580dc2ebafd9071184b8859e75ebeeb259a"}
{"id": "train_21854", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "evaluation_metrics", "auditability", "tooling"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0884f2bd3e5fcad9b5df27b24c1cd1ea4b457700"}
{"id": "train_21855", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "security_gates", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "032ffcfaa0a0b897fb3482124115135523e3ac93"}
{"id": "train_21856", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "governance", "reproducibility"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "27d0daf20cc1ce816a4e1cd541b2f6da95873bbd"}
{"id": "train_21857", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["governance", "tooling", "reproducibility", "auditability"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2fc6bd0d72794a4f472ae5781a093a9f49bf0e0f"}
{"id": "train_21858", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "tests_are_truth", "reproducibility", "documentation"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9dea90776c720a3335dc39ae8de4d75aa33927fa"}
{"id": "train_21859", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "ci_integration", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6e08f6538fe459adf36f00b7a59a6922213d9686"}
{"id": "train_21860", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "auditability", "governance", "ci_integration"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "80c93df682ce284933ad9eff2b6cd4d94ee9e0a1"}
{"id": "train_21861", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "evaluation_metrics", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "77b1e62ee94107974dd0c3b8c42a6268872e76c2"}
{"id": "train_21862", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "repo_scale_reasoning", "reproducibility", "documentation"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c3d884570ac6bce9e536b3f9347a3ff4b7a0596f"}
{"id": "train_21863", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["governance", "security_gates", "documentation", "tests_are_truth"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "458084efc9c41289077127267a478f75ad30ac24"}
{"id": "train_21864", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["governance", "repo_scale_reasoning", "ci_integration", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bc176671172e548e88bf5340039342611cc9374a"}
{"id": "train_21865", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "ci_integration", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0c83b29780bbc2e5f4c41f76898ba284506f9f59"}
{"id": "train_21866", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "tooling", "evaluation_metrics"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4fa409aa48f41599130a251565ae034fba93f6f7"}
{"id": "train_21867", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "auditability", "tooling", "tests_are_truth"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d722b93bbfe7c2d8cb5914a96508b5cbd11c1a47"}
{"id": "train_21868", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "tests_are_truth", "governance", "security_gates"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a547c060dd1a2b52527c96dc36ba889a291f698a"}
{"id": "train_21869", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "documentation", "reproducibility", "auditability"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dd542a6f8752a394242eacd1f5db614317f46ee5"}
{"id": "train_21870", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "governance", "auditability", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4c556ede5704a3468639f4a6023b0a3dfb2544fb"}
{"id": "train_21871", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "documentation", "repo_scale_reasoning", "governance"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "526211db4c366083cd4b243809f19c5d7976c722"}
{"id": "train_21872", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "evaluation_metrics", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5a9239f603172f422530eff83215e19bdcf1a7aa"}
{"id": "train_21873", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "tests_are_truth", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "13debddf7ff43169c5bcf5effed7b1b7e9ce1151"}
{"id": "train_21874", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["governance", "evaluation_metrics", "documentation", "tests_are_truth"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "eb909105026b425ca27fa9d98947ad41afa4acf2"}
{"id": "train_21875", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "documentation", "ci_integration", "security_gates"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "739c510011625d6ad0c068e738289443425b0cf1"}
{"id": "train_21876", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "documentation", "repo_scale_reasoning", "tooling"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6db4153a71d519cc68e8c5155965ada392a817c7"}
{"id": "train_21877", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "repo_scale_reasoning", "documentation", "security_gates"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fe12cc3492f49edb2ddc23dcbbcbb423c11d6230"}
{"id": "train_21878", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "governance", "tooling", "security_gates"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5f16089808f51828a332b797d5f7f45a44f1b32a"}
{"id": "train_21879", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "security_gates", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1d6c9c1eea7ad441d1540539df9af1a8ed27eeac"}
{"id": "train_21880", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "evaluation_metrics", "tests_are_truth", "ci_integration"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "36a372cde038ec737669f0f8fe070e15e51fe758"}
{"id": "train_21881", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "governance", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bd151c78b7da84bf2ff1cb3ec9e81b4a29857b1b"}
{"id": "train_21882", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "tooling", "documentation", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "623d62efaa5e0545d8fe4a928b5acb8deb31778d"}
{"id": "train_21883", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "auditability", "documentation", "tooling"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6776a724ec6a6758eb9d24ab1c003f511b8a2d66"}
{"id": "train_21884", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "repo_scale_reasoning", "reproducibility", "auditability"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f4ba27494e2d6b53be2ac5c9bd0b7d532091ca2a"}
{"id": "train_21885", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["governance", "security_gates", "repo_scale_reasoning", "auditability"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ca7c870de1b7b66802d01cecbfb018dc7bc69739"}
{"id": "train_21886", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "tests_are_truth", "tooling", "ci_integration"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3ba22e18d977021279f8d6d34270c188b6b72522"}
{"id": "train_21887", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "governance", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bfb0ae90860c5a0abc4c5d526bdac730214cd09b"}
{"id": "train_21888", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "documentation", "auditability", "governance"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6b3f5e4767abb852a271e2606949214fe501c111"}
{"id": "train_21889", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "ci_integration", "documentation", "reproducibility"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "03f0cecf2a1db2a39c7423de3c51f2bf74e7834e"}
{"id": "train_21890", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "tooling", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "efabbbd83d78802173707220214a43297a7d23bd"}
{"id": "train_21891", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "security_gates", "governance", "tests_are_truth"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c80790b8d538ebbe537a83e65f2df01df4a4dec0"}
{"id": "train_21892", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "ci_integration", "reproducibility", "security_gates"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "53b256c53d05e666131c3b93e1fb857a05136598"}
{"id": "train_21893", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "documentation", "repo_scale_reasoning", "governance"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f7c920d4b9076f3cdc69283a456cde4b4daf51eb"}
{"id": "train_21894", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "security_gates", "documentation", "auditability"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "96fe001e36bc7613f8909ada150d22c0aa6a3306"}
{"id": "train_21895", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "documentation", "reproducibility", "tests_are_truth"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ef257f193ef93ba6817239ad9d878c20be489346"}
{"id": "train_21896", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "tooling", "documentation", "auditability"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1000a0872705356d6ad708372a3550b86c3b0b3e"}
{"id": "train_21897", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "reproducibility", "tooling", "security_gates"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f9951c185210cb6e53c6bac087c9c6089a8ef37c"}
{"id": "train_21898", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "auditability", "repo_scale_reasoning", "evaluation_metrics"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1aa8fe4c1f39f8a7a0827e65b5a3a501d975e896"}
{"id": "train_21899", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "tests_are_truth", "ci_integration", "evaluation_metrics"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "830a9fd51f5b725eba8bce7cb84d79cc33201fd9"}
{"id": "train_21900", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "documentation", "governance", "reproducibility"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "805b8154faddbf1963a15a64433bf336d86e4ec7"}
{"id": "train_21901", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "documentation", "evaluation_metrics", "security_gates"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a62c81d864d9b9a59434b98b9b817ce56d21b361"}
{"id": "train_21902", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["auditability", "security_gates", "documentation", "governance"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "528175710d65be7374506190593b5ea4e41f6155"}
{"id": "train_21903", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "documentation", "security_gates", "tests_are_truth"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a47e1e0f5d79da33ebdf78100e552de9db382205"}
{"id": "train_21904", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "documentation", "ci_integration", "governance"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "83ad873fb4e3ef3ed8d776b659cd42b14a35aac4"}
{"id": "train_21905", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "documentation", "auditability", "security_gates"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "41f230656241a49b0e77e101fc14f4dcf6f2cf24"}
{"id": "train_21906", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "governance", "documentation"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "444a5436f207cc06a9831831f41e5bd194055f51"}
{"id": "train_21907", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["governance", "documentation", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "25ed7be5b30b06d54aab1e8d99441575517aab9c"}
{"id": "train_21908", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "ci_integration", "tests_are_truth", "auditability"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "78da504c7d0c219674bb80e114cbdfb3919b6575"}
{"id": "train_21909", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "governance", "tooling", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5ab5ff8c8094cc8fcd7f7adfd7c9b6eafb41dcb5"}
{"id": "train_21910", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "auditability", "security_gates"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d1be3c7876974f15ec968d0a6ab9b30f9e3e7499"}
{"id": "train_21911", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "governance", "evaluation_metrics", "auditability"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "764e89c29b808941da076134835c0f097ea23820"}
{"id": "train_21912", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "security_gates", "ci_integration"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "284ab27e7f3883f53170f1c6d3d4753a46afca58"}
{"id": "train_21913", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "evaluation_metrics", "tests_are_truth", "security_gates"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2cf3b640d28f27a3d0e1feff2db768e5d9efa6bb"}
{"id": "train_21914", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "evaluation_metrics", "auditability", "governance"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9bdff2309885852a13e64664af60ffaf19bfd17d"}
{"id": "train_21915", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "evaluation_metrics", "reproducibility", "ci_integration"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6e10ca851f24eb3468bb0eb8fd2903538741b7cf"}
{"id": "train_21916", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "auditability", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2f47b9a2c738c6ecb0b6e6bace2ce99e07fe0d9e"}
{"id": "train_21917", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "auditability", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "41d5964d25c10dc2982831af03b810b908076d9b"}
{"id": "train_21918", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "tooling", "governance", "documentation"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "75fccde20fc7b45756c0aebd6b85240d9279ed17"}
{"id": "train_21919", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "reproducibility", "evaluation_metrics"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6f73c12eb9de5fe0f3382055a85f2cc1ed9e2c22"}
{"id": "train_21920", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "evaluation_metrics", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "af3315f73e52318ae09784a957c85a36ffd0c4c2"}
{"id": "train_21921", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "governance", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4a8edc969e135cbbe381c1516598c9add456b07c"}
{"id": "train_21922", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "repo_scale_reasoning", "auditability", "documentation"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dbc782db23ef8469d7c803b5e75ec66279ac015e"}
{"id": "train_21923", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "ci_integration", "documentation", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "be7b18783578e9296bb8704342726f2a59f6e093"}
{"id": "train_21924", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "tooling", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f48cb39e0dba2e45b8216fce58250cf7a5437406"}
{"id": "train_21925", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "documentation", "tooling"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "650203d78b3dbf803ac5b132cceb8b28c8a17e58"}
{"id": "train_21926", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "security_gates", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bd52b598580c4a17e53548fd3b7482a69bbbb9b7"}
{"id": "train_21927", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "tooling", "reproducibility", "tests_are_truth"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "223e7781d4a7af911a27617028ea73b1840e1aaa"}
{"id": "train_21928", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "documentation", "ci_integration", "reproducibility"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0004acf31fb01a12ecd583dd4b4c7343c895d495"}
{"id": "train_21929", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "auditability", "tests_are_truth", "reproducibility"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "16f6e237461e76149fb67a9214e2b017edb309e3"}
{"id": "train_21930", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "ci_integration", "repo_scale_reasoning", "auditability"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4ee444e6e79708fac0690cb0a9e17300d37b5955"}
{"id": "train_21931", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "tests_are_truth", "security_gates", "tooling"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "59fefff0062fec7532a5f191e8051c57e5cf900f"}
{"id": "train_21932", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "evaluation_metrics", "governance", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b639c5452a90e7b11e47b50989f186eb7db8f68a"}
{"id": "train_21933", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "evaluation_metrics", "security_gates", "ci_integration"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8b4f36793dce867cd1085786fd67ddf3cf7158f8"}
{"id": "train_21934", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["governance", "evaluation_metrics", "tooling", "ci_integration"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bf5fc704a6d22ce8a96cbc6d69b6f93bdf905ae3"}
{"id": "train_21935", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "reproducibility", "documentation", "security_gates"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a21cee340fd2f874de6f9eb4a4f35cb97c4a7b3b"}
{"id": "train_21936", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "evaluation_metrics", "ci_integration", "documentation"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cd27d4cd801f7f2624e56829a8edfbad0802aff9"}
{"id": "train_21937", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "evaluation_metrics", "security_gates", "documentation"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "62a18985687a9b4d1c2e550392fc2c9fac8bf5a0"}
{"id": "train_21938", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "documentation", "evaluation_metrics", "reproducibility"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1626ca14dc64c6654bbc3cb6db34130d0512417c"}
{"id": "train_21939", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "tests_are_truth", "ci_integration"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "27e1bb0f2a9fe22df6ad6f499c0d3e173cda4ec0"}
{"id": "train_21940", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "cost_latency_tradeoffs", "tests_are_truth", "tooling"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "67d7007c6a4a069dfdc460a9bb9614a0eee59253"}
{"id": "train_21941", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "governance", "ci_integration", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1feb7eca9b21012c1c2f1e7f1f83a346a75b5535"}
{"id": "train_21942", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "tooling", "ci_integration", "evaluation_metrics"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2d4b663604e7f54b59f477f7baf8e87474aafe83"}
{"id": "train_21943", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "documentation", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8f85ab7e42b34059b912dc73abf93ee798022090"}
{"id": "train_21944", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "auditability", "ci_integration", "tooling"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9565440dbd6c245750596a137a40ebde11e6a617"}
{"id": "train_21945", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "reproducibility", "tooling", "evaluation_metrics"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ebe5b3fe380edbf2eda526c55e0dcf0f83161817"}
{"id": "train_21946", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "reproducibility", "tooling", "documentation"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "558c308f0ce2dc4e77515daefc75bab2cc3c6cf0"}
{"id": "train_21947", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "governance", "security_gates", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "aaa8cee05e6b37d3dcf41553cf719f39dc1e62a6"}
{"id": "train_21948", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "evaluation_metrics", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ba914c3b2dcb00a5b232282865bad9166307df4c"}
{"id": "train_21949", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "evaluation_metrics", "reproducibility", "documentation"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b25df752428455c15b5507f0b913f8e7b9290c6b"}
{"id": "train_21950", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "tests_are_truth", "documentation", "governance"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8373ead887dedaab081661ccde1fe18f83c82a69"}
{"id": "train_21951", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "governance", "tooling"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5eb90f0d8155ddedef97d8cae69b56aa4f0dd27d"}
{"id": "train_21952", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "evaluation_metrics", "cost_latency_tradeoffs", "governance"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2777281a6939c008f19082ee5d4d657b6b8b3028"}
{"id": "train_21953", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "security_gates", "documentation"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b994cedd70e0b935ae58528ee4a141014f88309b"}
{"id": "train_21954", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "auditability", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "417940e4112735d4fff954d4456deea806125fc4"}
{"id": "train_21955", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "governance", "auditability", "tooling"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3ca4f53bdace4a5d1e2641c28a0e0bb94f820531"}
{"id": "train_21956", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "governance", "tooling", "documentation"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3816932c8810379f948ffafd45527e1f18db769e"}
{"id": "train_21957", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "ci_integration", "tests_are_truth", "security_gates"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f00e57c5d7d39b00f879b53bce40a1483698d35a"}
{"id": "train_21958", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "reproducibility", "documentation"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5b9099022bef910d92e9c16430ff6a997773c5fd"}
{"id": "train_21959", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "tooling", "security_gates"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ddd02f85e5bcd42bf2e3dea1fcf6e4a9c60cf44d"}
{"id": "train_21960", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "ci_integration", "documentation", "auditability"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7c368f1cbf38cb216504c136de12407c647f233b"}
{"id": "train_21961", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "security_gates", "governance", "ci_integration"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f1ac4516355b783ef47604fe751bf67fa3f2559b"}
{"id": "train_21962", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "documentation", "tooling", "ci_integration"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2ab820109a35e446a6314d2063dc0a8ca3a19add"}
{"id": "train_21963", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "tooling", "evaluation_metrics"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4e3d21bb280821b8f865b506bb071f392fe1a227"}
{"id": "train_21964", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "auditability", "cost_latency_tradeoffs", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c03d1fefb52565bf398922c246e15d28473caeab"}
{"id": "train_21965", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "security_gates", "reproducibility", "ci_integration"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2202ce2bca1cac2cfb878d55aa34e41e03922670"}
{"id": "train_21966", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "repo_scale_reasoning", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5a83455abeb4f5021f645b8cc881a1af9ec08842"}
{"id": "train_21967", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "security_gates", "ci_integration", "reproducibility"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8d33947143dfd6263b379d7c2a41f89ffdd89fd0"}
{"id": "train_21968", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "tooling", "documentation", "ci_integration"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "66d3972c2210bd8e67b66276bd12d2f7b78607f6"}
{"id": "train_21969", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "governance", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "62df666d82a9df7dd3d5541e44446865a33cb326"}
{"id": "train_21970", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "cost_latency_tradeoffs", "governance"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d1da88f61f6ada844bf45332f0e8a31e3560d8b1"}
{"id": "train_21971", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "governance", "tooling", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "16757e9d406b0c1d8ca31107fd8fa741acff5874"}
{"id": "train_21972", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "documentation", "governance"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5216ce54def63becaae625c29c98523fb8db9390"}
{"id": "train_21973", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "auditability", "tooling"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bb9d9cae40681747591a3bd2421f64beeb4b24a3"}
{"id": "train_21974", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "tests_are_truth", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fe8d6543e889dd02b9ca3c9bffbdd0e9277ac1c3"}
{"id": "train_21975", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "reproducibility", "security_gates", "governance"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "df8c22acfd3ff5b991156c3360857f1e191d2314"}
{"id": "train_21976", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "evaluation_metrics", "governance", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b20fa4b21fa02486eb7283b479854a251989b8bb"}
{"id": "train_21977", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["governance", "tests_are_truth", "documentation", "ci_integration"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7a7cb1414d8fd1f7e8b2fb47bc4c487f57cd93bf"}
{"id": "train_21978", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "evaluation_metrics", "ci_integration", "security_gates"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "972f8ffe02b8cdc1a3c9e16a49d5166480f02aa4"}
{"id": "train_21979", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "documentation", "tooling"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bb2b60dc7026dca12c4845dcb9cb8964626264b6"}
{"id": "train_21980", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "governance", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b92b7ff8982c5e7dbdff2d9f34e72e609bf7c8ef"}
{"id": "train_21981", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "tooling", "tests_are_truth"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2ab5167126159f012fcbf45c441fb6ac1b24ffe8"}
{"id": "train_21982", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "tooling", "auditability", "documentation"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9066ed9e134395ae20289263f3d65fe584dc031f"}
{"id": "train_21983", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "auditability", "ci_integration", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4e4d3967cc6a3199a3f616fe764f2c6c75a96850"}
{"id": "train_21984", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "repo_scale_reasoning", "tooling", "security_gates"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ea4e116f17623d329d2fd6b42da6a746c6123006"}
{"id": "train_21985", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "documentation", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5d54a6ca7b3906cd37184ce18d2f6faabc316aa1"}
{"id": "train_21986", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "documentation", "tooling", "auditability"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bd2b7927a8e9e8c49975138dec88b965fba4ad50"}
{"id": "train_21987", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "security_gates", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bc5f2c6d34388ec09fd0a7c2ededd76cb17eef7f"}
{"id": "train_21988", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "auditability", "tests_are_truth", "ci_integration"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4856d236b6d7e059bcb005acd5d4279e63d1d1cb"}
{"id": "train_21989", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "documentation", "reproducibility", "auditability"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "14752f0f268002b0dafa5792fc74d921962e7f04"}
{"id": "train_21990", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "governance", "evaluation_metrics", "reproducibility"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4cb2c3c10e020f3c42b270a6f4f4f67f2736b1b8"}
{"id": "train_21991", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "documentation", "ci_integration", "repo_scale_reasoning"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8ab131190ef3ab4e8336ef8beb37ea4e4df6db58"}
{"id": "train_21992", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "tests_are_truth", "repo_scale_reasoning", "documentation"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c3d6d4c55051272796efd00d527fc2f6fb19f6de"}
{"id": "train_21993", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "repo_scale_reasoning", "auditability", "governance"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e0cbc00ea784253fa2892fbfa98674092f6226ff"}
{"id": "train_21994", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "security_gates", "repo_scale_reasoning", "tooling"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "040641173cd03d3517c46b817f9466f86324b18e"}
{"id": "train_21995", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "security_gates", "ci_integration", "documentation"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e257b5d4a83d9a61a30a5b00042ab3ea79cd0c3e"}
{"id": "train_21996", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "security_gates", "documentation", "ci_integration"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4775176e1e6c658fb6d6579518322c90fd85f636"}
{"id": "train_21997", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "reproducibility", "tests_are_truth", "auditability"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "00a8041795e2cac337bd80fa667df27b1a678134"}
{"id": "train_21998", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "ci_integration", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "42b5cd24e946832da2cdcc6e3170e5c4851602a6"}
{"id": "train_21999", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["governance", "tests_are_truth", "repo_scale_reasoning", "security_gates"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9a8f57acf724c356e1f0627bc59043c61d15e7aa"}
{"id": "train_22000", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "reproducibility", "documentation", "ci_integration"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "123a195a40e7c7159df7daba3511ba04251982e0"}
{"id": "train_22001", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "tests_are_truth", "ci_integration", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "580a39b7191b2df3db058658581bf7336bc18ef4"}
{"id": "train_22002", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "repo_scale_reasoning", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "81667b547f60eaf54634a4671061f651e407be5d"}
{"id": "train_22003", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "reproducibility", "ci_integration", "governance"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a0ae5631d4de2f3c70d59f93ebe52a99b1be4ebb"}
{"id": "train_22004", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "governance", "ci_integration", "tests_are_truth"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b04d5226090ea7cde1d60550632afa048db924d2"}
{"id": "train_22005", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "reproducibility", "ci_integration", "security_gates"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1a3900c523ed4663420092dd911d3b0201db7a63"}
{"id": "train_22006", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "governance", "repo_scale_reasoning", "tooling"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "30cb6478a7424171135037e0430aea8014a6ef77"}
{"id": "train_22007", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["governance", "ci_integration", "cost_latency_tradeoffs", "security_gates"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6bf819ba8871a82b9c0ea5738d0323a1cdfebe6c"}
{"id": "train_22008", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "security_gates", "governance", "tests_are_truth"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "09a79990b2e8019383c5838350ed2c2e79a8648b"}
{"id": "train_22009", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "ci_integration", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "148ebcc3a074349e4e505bc3b92f9d7c1df1861b"}
{"id": "train_22010", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "tooling", "security_gates", "reproducibility"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2d01672dd74c62b90ebd1fff46fe25059f615c72"}
{"id": "train_22011", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "repo_scale_reasoning", "cost_latency_tradeoffs", "governance"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f46441a94792cb47d69fc6c44a760559da5a552d"}
{"id": "train_22012", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "repo_scale_reasoning", "auditability", "documentation"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "49d9b416ec50499a7389f72c41bfaba6868b0693"}
{"id": "train_22013", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "auditability", "governance"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c79a42a97fae5292787d9ee678caf5eb2910dd59"}
{"id": "train_22014", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "auditability", "ci_integration", "security_gates"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3cd61dcde8ddac90c9012b716e630b6aef068dea"}
{"id": "train_22015", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "reproducibility", "ci_integration", "tooling"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3a6bab8c101b13f4e44cd2cde82d01ceead69df1"}
{"id": "train_22016", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "ci_integration", "governance", "auditability"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "175bc2ce9dc2135c5d7dfc19440dc0f178d45785"}
{"id": "train_22017", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "tests_are_truth", "ci_integration", "documentation"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d37c7a0b97b2c38f10fe6d17795fc960495b7484"}
{"id": "train_22018", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "governance", "tooling"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6ac5ba4b00cd65cb0cbd2247a9759acbee8d2123"}
{"id": "train_22019", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "security_gates", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0a1534feee8c814f463352c76ecfbf5ac67ba2fb"}
{"id": "train_22020", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "governance", "tooling", "reproducibility"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6d704c5a7ae82d55e9d6f265788580cb9b731ba5"}
{"id": "train_22021", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "reproducibility", "security_gates", "tooling"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "231acb947631deb60c30c4f32c47008fa93a0c65"}
{"id": "train_22022", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "governance", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dd5ce1e9943ef5a557e64de4ba40b5c6799eca86"}
{"id": "train_22023", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "documentation", "security_gates", "auditability"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1f537a09573328f91cea9118c14d70b423deb364"}
{"id": "train_22024", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "documentation", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bce0af83e3011f49d8a0421e7a8ba1cf161b41b3"}
{"id": "train_22025", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "evaluation_metrics", "repo_scale_reasoning", "auditability"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1e6b0558eb46448f3ef27b33fee6449550dc81d2"}
{"id": "train_22026", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "auditability", "evaluation_metrics", "reproducibility"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2b6ff48032b31eb606500509d562e5be077674df"}
{"id": "train_22027", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "reproducibility", "ci_integration", "security_gates"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c4771fa14b370eef8bdc3d3300c3afae49476b65"}
{"id": "train_22028", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["documentation", "repo_scale_reasoning", "tooling", "auditability"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4eda554f69644ffb3face177d25bd3b4e0011d5c"}
{"id": "train_22029", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "documentation", "reproducibility"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "28d170fe0324917376e1b92f7cfd776bf5ae0ce7"}
{"id": "train_22030", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "evaluation_metrics", "tooling", "governance"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "427547d489a9f01997fd214e6b7fba41de4d4814"}
{"id": "train_22031", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "documentation", "reproducibility", "auditability"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "893ea222cb98d3c7de447c2c61457944a418dcd0"}
{"id": "train_22032", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2556f98fbe06ef53a41e878a9a7599d69ed16f63"}
{"id": "train_22033", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "documentation", "ci_integration", "auditability"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4cc4d086d5c4be76a448824dafc9ab00b19df2c8"}
{"id": "train_22034", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "tooling", "ci_integration", "tests_are_truth"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f2e572b9b8493cb2e1dde6c7c227d483e185d4bd"}
{"id": "train_22035", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "repo_scale_reasoning", "tooling", "reproducibility"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3fc58c7db701dc1d0f0be881fd2b987076df213c"}
{"id": "train_22036", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "security_gates", "ci_integration"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "447e391e2740386379d43da6cc21cb92fad73aa2"}
{"id": "train_22037", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "tests_are_truth", "tooling", "security_gates"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "160f14673dd3353ede01c043efadc978c2f625fb"}
{"id": "train_22038", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "reproducibility", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dd3dba469a31412ff0f31a35f351587d41b2ce2c"}
{"id": "train_22039", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "security_gates", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9fffcfe5c1008244efa216da47949ac6edae7e8d"}
{"id": "train_22040", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "reproducibility", "repo_scale_reasoning", "tooling"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f9ebd6e5c7970b287eb4d7a24f969c437cc0c2d7"}
{"id": "train_22041", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "tooling", "auditability", "evaluation_metrics"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3f80d7373d273fbab80b7aace57b44d2d5056bec"}
{"id": "train_22042", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "reproducibility", "auditability", "governance"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c8bc6011b662d52628e7661025dab7c7d3062bba"}
{"id": "train_22043", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "security_gates", "governance", "tests_are_truth"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "03ebc0b3319b2ec6affcb203b245e00482b6d1f0"}
{"id": "train_22044", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "cost_latency_tradeoffs", "evaluation_metrics", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bdb9a50a304db8029ba335bd7b18afa0ffe4acc4"}
{"id": "train_22045", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "tooling", "tests_are_truth", "governance"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "369d4d9c832aa4635b0aee352ab6e132e9988f27"}
{"id": "train_22046", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "evaluation_metrics", "documentation"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "851491f0b7253cc9457031cb756427de7fd88c12"}
{"id": "train_22047", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "cost_latency_tradeoffs", "ci_integration", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c8caac8d771bcd07c88ec9490c6f62d01a2eb34d"}
{"id": "train_22048", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "ci_integration", "tests_are_truth", "documentation"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c4e1deb77a39322050ed9d617cf3995427c4fa42"}
{"id": "train_22049", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "governance", "auditability"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c7f31c36d34e33ad722680004d4d85babc440d18"}
{"id": "train_22050", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "security_gates", "documentation", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "64a03cb06b4f8fc28bf1aceba6a0e76a964f59b1"}
{"id": "train_22051", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "security_gates", "tooling", "tests_are_truth"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "53ccc119b3cde2e71cc00feaec48cb91d8d870c2"}
{"id": "train_22052", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "tests_are_truth", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "41b701e43b6267f3f20d18d8f5a66082b852edd2"}
{"id": "train_22053", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "auditability", "documentation", "tests_are_truth"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7d13a413377240baaf064bbb972eaca22268c175"}
{"id": "train_22054", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "evaluation_metrics", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "44b84ac93e0855fe899ba18f7e7cd3a0dc3813fe"}
{"id": "train_22055", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "auditability", "governance", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ab79efa2a9b406afbbc76ce74b93ded7626b4019"}
{"id": "train_22056", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "documentation", "tooling"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b986ab7daa8e996fd2bbe0b249bef809f97a5bcd"}
{"id": "train_22057", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "documentation", "repo_scale_reasoning", "tooling"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "98c5cedec94520c4a572f94e33ac9d04b12f0615"}
{"id": "train_22058", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "repo_scale_reasoning", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0c93384e101bf5d1c4c466568ed44af1284d9d33"}
{"id": "train_22059", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "documentation", "security_gates", "evaluation_metrics"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "af4fa1a2d6c7a396907c41069674ef17fe9a738c"}
{"id": "train_22060", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "evaluation_metrics", "documentation", "ci_integration"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cc1a7a6093780c6ce6b826cf7f99bbf4eabc2d30"}
{"id": "train_22061", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "auditability", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f9ec36b79e2d543d3ef09fd89b37b3969e15d8f4"}
{"id": "train_22062", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "evaluation_metrics", "ci_integration", "repo_scale_reasoning"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "95bff4e872c7d2905f702e71d22ff3ce90db22f4"}
{"id": "train_22063", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "reproducibility", "security_gates"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b928f7171c524f2097743553f79f6898b6c68612"}
{"id": "train_22064", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "governance", "auditability", "security_gates"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ed1c7c0d8e9a60a59114aeaa7097e6fce4d8ccb1"}
{"id": "train_22065", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "evaluation_metrics", "cost_latency_tradeoffs", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f621e30bd75e0958c61a42d5a4b8627f8c1219be"}
{"id": "train_22066", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "governance", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "589cea79f06b30bf029e112cb4d7d27d77c646c4"}
{"id": "train_22067", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "auditability", "documentation", "security_gates"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "49c3a2f6c32ecdc9f7264fc3afffc68b638f72b2"}
{"id": "train_22068", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "evaluation_metrics", "reproducibility", "governance"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "57668620491d6ad5394047736fba658dd48007e0"}
{"id": "train_22069", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "evaluation_metrics", "auditability", "repo_scale_reasoning"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "16ee4b6aa4333295046bc13a9fc64d0605c61a46"}
{"id": "train_22070", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "evaluation_metrics", "cost_latency_tradeoffs", "governance"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2ca73aa0ff6ec5b240b1048aefdda2c6a77265fd"}
{"id": "train_22071", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "ci_integration", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "71b380e9f983fc4ffb8a844abb879429ce0f9225"}
{"id": "train_22072", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "reproducibility", "tests_are_truth"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2abdb85fa291a36442a74c8c4b5dfcc1f7c68905"}
{"id": "train_22073", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "governance", "documentation", "tests_are_truth"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0008fabc5a6c94e38caf77c3fbbfe5fa13768973"}
{"id": "train_22074", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "tooling", "reproducibility", "documentation"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "258bce57b732aeb3ca53ebee74c8168a8cdf699d"}
{"id": "train_22075", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "governance", "evaluation_metrics", "ci_integration"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9567939b06505b735380a5b142c923de3e949008"}
{"id": "train_22076", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["documentation", "security_gates", "auditability", "tests_are_truth"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1441c144537dfd5a24a4a8deafea443fd4c61360"}
{"id": "train_22077", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "governance", "tooling"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a6f1679ace860b7a2ebfd00625f65d40c033aa7a"}
{"id": "train_22078", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "auditability", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0c133dba480b6e8ae6ae04dae78f39f23f112c02"}
{"id": "train_22079", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "tooling", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "78ff4a8de894b647a07cec19e18b6a452daa0f71"}
{"id": "train_22080", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "ci_integration", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3a237e6c61b8eeb575c973d576e3d5eac955ebce"}
{"id": "train_22081", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "ci_integration", "tests_are_truth", "documentation"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2f35d5e99fa3e6dee811487a7c3949f43e961516"}
{"id": "train_22082", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "governance", "repo_scale_reasoning", "security_gates"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d91cbdefc98856e5672e54a701d72ad2b79a3616"}
{"id": "train_22083", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["documentation", "tooling", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1436857db7dcbf4c855f60dce6f247f1bf9f1f0c"}
{"id": "train_22084", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["auditability", "evaluation_metrics", "documentation", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "539ef4ca0ada37e2884086094bd75805ed59ea04"}
{"id": "train_22085", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "documentation", "governance", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c1099d451e8958a3fbdffe50de9c7088a0df86c0"}
{"id": "train_22086", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["governance", "tests_are_truth", "security_gates", "auditability"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "70d338fe4b37b7cefaa869a3654bec1b0a7ee9e9"}
{"id": "train_22087", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "governance", "evaluation_metrics"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0a1238a2284adfafd106905f3a1e469042eed720"}
{"id": "train_22088", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "documentation", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d05f40f0beab816ee38ae6118bdb8ed82c99ee8a"}
{"id": "train_22089", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "auditability", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "87061aa3e6b9f8ce79af5ebfddee670ccff41813"}
{"id": "train_22090", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "tooling", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "77e2c918718a203a55fb55006011c193bd95622f"}
{"id": "train_22091", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "documentation", "repo_scale_reasoning", "tooling"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4d93c9a556a8809c750426d97d9c05c49730a209"}
{"id": "train_22092", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "ci_integration", "governance", "reproducibility"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "508f3fd22b541bef3b2b859099899fc3ae619d59"}
{"id": "train_22093", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "reproducibility", "tooling", "tests_are_truth"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "314ed10ff5485d1b4a7945291aa669e4608bc3f4"}
{"id": "train_22094", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "reproducibility", "tooling", "ci_integration"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f0c4885e2dd8655d3ffe63947b0ed99309bca556"}
{"id": "train_22095", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "auditability", "ci_integration", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0d8fe929e9d185977c6671a8c8687f5313e0a369"}
{"id": "train_22096", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "governance", "tooling", "auditability"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a9f0b6172b744634768151c92c1bd76f65e30328"}
{"id": "train_22097", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "security_gates", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2d2ca73cf706e7d0ce27559088d4a9d0376e8609"}
{"id": "train_22098", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "tests_are_truth", "repo_scale_reasoning", "auditability"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9d1df00035210c4253cf8722595886b6c878d3fc"}
{"id": "train_22099", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "reproducibility", "ci_integration", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "097fdd55ccf4a444eabe77b9a86cd73bf76c6c92"}
{"id": "train_22100", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "tests_are_truth", "documentation", "governance"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5d6ddcb3bf19a13f8cd102fee0d2c7e6e60d3847"}
{"id": "train_22101", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "auditability", "cost_latency_tradeoffs", "security_gates"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c38a13760bb50ee6a8532352f4d5d4baa2a82130"}
{"id": "train_22102", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "governance", "tests_are_truth", "ci_integration"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b243227be1d35665108498112b1dd023849ca01b"}
{"id": "train_22103", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "evaluation_metrics", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0ba2ef7d41ab9cf2b0f8ac2246f036b713d4d036"}
{"id": "train_22104", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "documentation", "ci_integration", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f15510cad5c70a5483de72ccb9f780645175b069"}
{"id": "train_22105", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "tooling", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "49ce12f46d7407c0d9f6a15ec0cb3f784be09ca8"}
{"id": "train_22106", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "reproducibility", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5c3830b344df9640b47143c950c9d26da72cb700"}
{"id": "train_22107", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "security_gates", "repo_scale_reasoning", "reproducibility"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "533d4e0745ecd454008377e2b4886665df532714"}
{"id": "train_22108", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "documentation", "ci_integration", "tests_are_truth"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c9374dd4b3a94492eecec75a00b5c481c6e45d6d"}
{"id": "train_22109", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "tests_are_truth", "auditability", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a4d8ae69a0653a19caf2342e7eb2c099edaccece"}
{"id": "train_22110", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "auditability", "ci_integration", "tooling"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9a04585f175c0a222a96e62d27641f759360714e"}
{"id": "train_22111", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "tests_are_truth", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e2531974312a81dbe7f6475d873732f4ff049349"}
{"id": "train_22112", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "security_gates", "ci_integration", "auditability"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "00e6a6c0eede6038c05876621b8b0daecc710d5f"}
{"id": "train_22113", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "auditability", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "089f19bb5310d569243c25a4483c4bf0b60aaee4"}
{"id": "train_22114", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "repo_scale_reasoning", "tooling", "documentation"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2eb41c5bcc084abf88c072bf18a095c7309390be"}
{"id": "train_22115", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "ci_integration", "documentation", "evaluation_metrics"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "88c77f03e83c08fd5ac45e557059ae64b87e888c"}
{"id": "train_22116", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "documentation", "cost_latency_tradeoffs", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4fae9208a6b3c788a6db68485c27ad253eef15e1"}
{"id": "train_22117", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "reproducibility", "tooling", "ci_integration"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "eb622528085ff0131c5a9a87d726b98a3c9ae79a"}
{"id": "train_22118", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "auditability", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c4eab1281cb5540c3a86cd54193092f2b43cf326"}
{"id": "train_22119", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "repo_scale_reasoning", "cost_latency_tradeoffs", "security_gates"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f41a5f421b70102bccc76817fae5fc9507a69327"}
{"id": "train_22120", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "security_gates", "evaluation_metrics", "tooling"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ab06bdb360d7a6a312ec3b8ba6dc9ddc5f9d00b6"}
{"id": "train_22121", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "reproducibility", "repo_scale_reasoning", "documentation"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ce6b4367462349d8e780a6118fd175a0c141f73b"}
{"id": "train_22122", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "tests_are_truth", "governance"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "992655cce64f5bb11da52fd387d1178f25ce1f9a"}
{"id": "train_22123", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1288a46a21a07eb536bb5cd66905623371b47275"}
{"id": "train_22124", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "ci_integration", "tooling"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d0f3d96e7085a1a113930b42ba1bf7407a30300b"}
{"id": "train_22125", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "auditability", "tooling", "reproducibility"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "545a35cc2ff0ca9ead19caf8e1f94ee293360826"}
{"id": "train_22126", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "security_gates", "ci_integration"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "64eeeadb6d0b6854cf0e28c351d4ead875d98049"}
{"id": "train_22127", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "auditability", "ci_integration", "documentation"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f9e90124c6b7a918920e80029f01308d9034f936"}
{"id": "train_22128", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "auditability", "ci_integration"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "32e13b46328c4dfcfdd6a26e4f2915d1e8846b0b"}
{"id": "train_22129", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "repo_scale_reasoning", "evaluation_metrics", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7f82b4a683ebbf15b09cf8d20104c4b7d55c9646"}
{"id": "train_22130", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "governance", "repo_scale_reasoning", "reproducibility"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b2f85a69bee67c048405e37c6475f1fd07da85fc"}
{"id": "train_22131", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "documentation", "tests_are_truth", "governance"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ed1a8cb9098d607f5ad6d5ffaab9286878b5c071"}
{"id": "train_22132", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "repo_scale_reasoning", "governance"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4e6f602ce5c6188f8ec037eb8327438c827741e7"}
{"id": "train_22133", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "security_gates", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "482d9fd377020a7991c1c69ed74657e0ab6170d4"}
{"id": "train_22134", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "ci_integration", "documentation", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5e8df88610e8112604d200feae7a67f9c9ff1f32"}
{"id": "train_22135", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "evaluation_metrics", "ci_integration", "auditability"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "40b99b7ee917e682b922390d8e1eccb79a39f1d4"}
{"id": "train_22136", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "documentation", "reproducibility", "auditability"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e409b9d61cbf8a3d9eb163be6f1e4e8a466c593e"}
{"id": "train_22137", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2671f58033cae575d4cbeee8a887e07f33cd1d5d"}
{"id": "train_22138", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "tests_are_truth", "auditability"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0033e5e147835f5589212bde897faa08d6b74875"}
{"id": "train_22139", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "reproducibility", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cf9fa46914c2bc2c4be469370a449253497be01c"}
{"id": "train_22140", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "security_gates", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a58bb3cabebe3cb5ceca156ce2b224930fc21d27"}
{"id": "train_22141", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "repo_scale_reasoning", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f75fa30ca3c7c6babbf0583c2384cee1f17f5c96"}
{"id": "train_22142", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "repo_scale_reasoning", "tests_are_truth", "ci_integration"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3cd5b0999f27497962e5e4d23d9682a1e2f84583"}
{"id": "train_22143", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "documentation", "governance", "auditability"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "670e009fb0387718d23887d88c8e90241aab47a7"}
{"id": "train_22144", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "tests_are_truth", "governance", "repo_scale_reasoning"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4ce272bb10473962673aaca020b48ca4c0ddb6a7"}
{"id": "train_22145", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "reproducibility", "repo_scale_reasoning", "tooling"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6c5d3bbed6336094efd247f4408fbd941f5c1b9f"}
{"id": "train_22146", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "security_gates", "tooling", "auditability"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "eb01fe8d083f561e79c322329f3e1df7006dc309"}
{"id": "train_22147", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "evaluation_metrics", "tooling", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b549417db1ae5bd501c8f07456422f7a2841d5de"}
{"id": "train_22148", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "evaluation_metrics", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ac02e9d98b4d56d67d14a0b38ecc87957438e1e9"}
{"id": "train_22149", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "ci_integration", "tests_are_truth", "tooling"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "01c101858cb3cc1b6965d52c607490556b671b2b"}
{"id": "train_22150", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "documentation", "ci_integration"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d111a7dc50cb88f38f3ea60783671fc1ef765243"}
{"id": "train_22151", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "security_gates", "ci_integration", "documentation"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a22e4eb6ed7b9b93defdf7a9d8395160c761c0a1"}
{"id": "train_22152", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "ci_integration", "tooling", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d7c799196d99d8e2d50262adb0993ed203c15623"}
{"id": "train_22153", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "documentation", "auditability", "reproducibility"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "32b8dcd01d3bf48a874d8cc332552d35b497b80d"}
{"id": "train_22154", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "tooling", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "afb05e99f5e9b7badb7a14993db076eb8bd1d622"}
{"id": "train_22155", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "governance", "repo_scale_reasoning", "security_gates"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fd7c8c753efd7360856465252e9dac6eaa950753"}
{"id": "train_22156", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "tests_are_truth", "auditability", "governance"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0ef684a3bc165e5a59b17660f590abd787822f3d"}
{"id": "train_22157", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "repo_scale_reasoning", "tests_are_truth", "governance"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "47aa6254e83e404937ddd6dbe4c9dbd60d9fd790"}
{"id": "train_22158", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "ci_integration", "repo_scale_reasoning", "governance"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4ced659436272de600aa54672950613a661804ff"}
{"id": "train_22159", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "auditability", "repo_scale_reasoning", "security_gates"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fa5f90c13401ef09a5a72d3aa6f4b456fa44aae0"}
{"id": "train_22160", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "reproducibility", "documentation"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d4c885860605ba3a9101b0d4539d861a2fc487ac"}
{"id": "train_22161", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "documentation", "repo_scale_reasoning", "security_gates"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d3c55836b5e7d8ad6c524a7350b534d6caf67df4"}
{"id": "train_22162", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "ci_integration", "tests_are_truth", "documentation"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f7be5ab949b3a69151a250bc54d31a794e5e3ca3"}
{"id": "train_22163", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "reproducibility", "documentation"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a532e82deab8cf4ff2e0aa582ea2349a77539824"}
{"id": "train_22164", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "tests_are_truth", "governance", "security_gates"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e5c0266463166559aa6f2587526d9a9a06450464"}
{"id": "train_22165", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "evaluation_metrics", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1a713524317b97115197d8f1e09946ed95969bfc"}
{"id": "train_22166", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "tooling", "cost_latency_tradeoffs", "governance"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "83ad873fb4e3ef3ed8d776b659cd42b14a35aac4"}
{"id": "train_22167", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "governance", "reproducibility", "tests_are_truth"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "48bcbd8c8999f54fa649a62e898d480932b3cbe3"}
{"id": "train_22168", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "evaluation_metrics", "ci_integration", "reproducibility"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8c6d0f24f888c9d234d5fb392579c0d0c9df2548"}
{"id": "train_22169", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "repo_scale_reasoning", "evaluation_metrics", "tests_are_truth"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "56b85187047e61756a1e97302b4a02d83b5d8513"}
{"id": "train_22170", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "governance", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "42c5de4e1851ffb82df89d9e2912d8d693a45c6a"}
{"id": "train_22171", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "documentation", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bae2c5198fa5d0a78a873024aab2ef5ddd852b08"}
{"id": "train_22172", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "ci_integration", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b76a148e23237b63d2e9d6088429c9a20b7068c8"}
{"id": "train_22173", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "reproducibility", "tests_are_truth", "auditability"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1287cee38742f0356328851077c2f99a0a8c10b2"}
{"id": "train_22174", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "security_gates", "tests_are_truth", "auditability"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b462615253db9270469e268b1c93da9cf21eb834"}
{"id": "train_22175", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "ci_integration", "documentation", "security_gates"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fb0c6b9988fccb9365b3dfecc71c7bb857043e8e"}
{"id": "train_22176", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "ci_integration", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "23acaf8db4a2174f68dd30346ccfda545e19bb4d"}
{"id": "train_22177", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "tooling", "governance", "ci_integration"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "99abcc1cac230846fb44439884e10af0c254d604"}
{"id": "train_22178", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "security_gates", "tooling", "tests_are_truth"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6b464b89e9944f0f157397f195eda20614f07e62"}
{"id": "train_22179", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "tooling", "documentation", "governance"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5fd8dbf78345694a5f715f65a460540746b0e93f"}
{"id": "train_22180", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "governance", "ci_integration"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "806f061d05cff75d46605638ab923c2a33f5a0bc"}
{"id": "train_22181", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a5811a87d1693bcb8e9441aeb5b482d390c27566"}
{"id": "train_22182", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "tooling", "governance", "reproducibility"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "42bd8bb4bc2e7b7603f92a81f82771f547097355"}
{"id": "train_22183", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "tests_are_truth", "security_gates", "evaluation_metrics"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "53573d774b5c7fd896816a86472efa4081ad73d2"}
{"id": "train_22184", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "evaluation_metrics", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "46775619661df0e051f30ff4f4f5717cd5ea5f5e"}
{"id": "train_22185", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "tooling", "documentation", "ci_integration"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "692a2f379fbd810d22d5093d55fc18c012c01815"}
{"id": "train_22186", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "repo_scale_reasoning", "tooling", "ci_integration"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d197f1538fb1ae8b86a9bedeb088101b12c22507"}
{"id": "train_22187", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "tooling", "tests_are_truth", "ci_integration"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "23f150093184637cf60f4de287b50e8501c4f051"}
{"id": "train_22188", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "tooling", "governance", "ci_integration"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b3a08f1dd50f3c377942a20314433fcfbabdb19f"}
{"id": "train_22189", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "documentation", "auditability"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8ba751efd2cc7ef6c3ee460e437bf13b6b5f8656"}
{"id": "train_22190", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "auditability", "documentation", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3114aace7c00041594bf28407d9f3b107a44f87c"}
{"id": "train_22191", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "reproducibility", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d5aaf3102de9084f968034ac30169e4badcf921b"}
{"id": "train_22192", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "governance", "tooling", "tests_are_truth"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3af7ecf7c569fe65c824205cfa14158744255790"}
{"id": "train_22193", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "documentation", "repo_scale_reasoning", "security_gates"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "899e6675c618b51c39e8841edaca0caa4b764a5a"}
{"id": "train_22194", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "auditability", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "09f5c6c684ee62d779313411de6d54faaa9673e9"}
{"id": "train_22195", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "ci_integration", "evaluation_metrics", "reproducibility"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9648fe749ce74968945762ac7559a00b68e1bf73"}
{"id": "train_22196", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "tests_are_truth", "security_gates", "reproducibility"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "57779c729ffa73c8f8bb18464ebce814f68c594d"}
{"id": "train_22197", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "governance", "tests_are_truth", "tooling"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "118c5013e7886f32c7f2f82977ca39d14239da49"}
{"id": "train_22198", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "security_gates", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4121a1658cb75d5ed546151ff095703120447b01"}
{"id": "train_22199", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "evaluation_metrics", "ci_integration", "tests_are_truth"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1e881a405a0217a7857e191e4894622ce1ee6073"}
{"id": "train_22200", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "auditability", "tests_are_truth", "reproducibility"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "021c7480b31533d9032c716820212d77e5b8a09b"}
{"id": "train_22201", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "auditability", "governance", "tooling"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "404fd19ef3aed4fc21b96f872885d46734d6cd8f"}
{"id": "train_22202", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "tests_are_truth", "ci_integration", "evaluation_metrics"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "612290fd7cece74b9dc0414165b78e3cd8b924d6"}
{"id": "train_22203", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "auditability", "governance", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f77c1293d6be316bcbdb2ea4a12cced406928065"}
{"id": "train_22204", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "evaluation_metrics", "governance"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ee9c2efa0cad28cc1c8d8036bcfc956254ce75ce"}
{"id": "train_22205", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "documentation", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cbd69dff66da5c4ca3bb87beea7b45a854349276"}
{"id": "train_22206", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "security_gates", "evaluation_metrics", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d42162a837ea62d0db6a8b2be46cd6433535f099"}
{"id": "train_22207", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "reproducibility", "governance", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6d607d36c02fa07605d620667e756a70b46945f9"}
{"id": "train_22208", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "tooling", "tests_are_truth", "governance"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "08ddaee0f4ae2db5a7058ca2f3ae0f246e45404e"}
{"id": "train_22209", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "repo_scale_reasoning", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8bef0bf1b2dc941a52195b3ad7f3d1db4b11bd5a"}
{"id": "train_22210", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "documentation", "governance", "tooling"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "53510ee5ab4ab8d1971cf0e0839b7aedf9a6ff49"}
{"id": "train_22211", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "ci_integration", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "134bde94580531c1e8cab356890a9d916c4b42d3"}
{"id": "train_22212", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "security_gates", "documentation", "tests_are_truth"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "85ff536d3f5b2bbcfce3796b57596b5c50684016"}
{"id": "train_22213", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "security_gates", "tooling", "evaluation_metrics"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7ec2600af037c7f4460a1e436cc2db76855c14c8"}
{"id": "train_22214", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "tooling", "reproducibility", "ci_integration"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6f7defb9e3f1ec2c90b39753d1eb7008511d39d5"}
{"id": "train_22215", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "repo_scale_reasoning", "documentation", "reproducibility"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b3671cee2db56b6d1a9475de2de4bf8bcb98c136"}
{"id": "train_22216", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "ci_integration", "security_gates", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3cda4946a1c65a1aff9f3f8319b938b65c05550b"}
{"id": "train_22217", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "reproducibility", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e793ccaf5a81ff7d1832cad6f7d76b43e6a8c776"}
{"id": "train_22218", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "auditability", "tooling", "documentation"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "80c077d6df94f9f66a2bee02f8b4e04b2973a480"}
{"id": "train_22219", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "tooling", "security_gates"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "75c8e4f075fbfa174887f3620a5e2a113bc5ef03"}
{"id": "train_22220", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "reproducibility", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fa949f6799d94773ff16ec92ccbc3f1b47ceeb25"}
{"id": "train_22221", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "governance", "documentation"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ed5b67006c263895598a98047b34118d2af7839f"}
{"id": "train_22222", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "reproducibility", "tooling"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7d955525b9fb54b31569e7973dcb09bd01b8f81c"}
{"id": "train_22223", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "auditability", "governance"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "92836de901b0b72e00dbd20823751882dc488686"}
{"id": "train_22224", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "security_gates", "governance", "tooling"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5234ceeea4c3ec6106795333409e4eb033dde46a"}
{"id": "train_22225", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "reproducibility", "security_gates", "ci_integration"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a4dbf7b19c14c99197d85c7fd7870968237afddc"}
{"id": "train_22226", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "documentation", "evaluation_metrics"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6ef1a6ae8dbe1c6ea7e3556b8d1a1153684ef092"}
{"id": "train_22227", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "documentation", "tooling", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "82061c3b512ad5604af7185008ed8b85cfe7e12e"}
{"id": "train_22228", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "ci_integration", "auditability", "security_gates"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f6fb37a3e7403955dbd3c57fba503503bb0a56a5"}
{"id": "train_22229", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "tooling", "cost_latency_tradeoffs", "governance"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "83cdd776448499657509bc3c9fe6f560c62bd2be"}
{"id": "train_22230", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "documentation", "governance", "ci_integration"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "40ff61dde4bff7fff8199d772a233596cf011e28"}
{"id": "train_22231", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "governance", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "611c681549d82a0f0fe822c856f32c2bfd599f80"}
{"id": "train_22232", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "reproducibility", "repo_scale_reasoning", "tooling"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d568b07f9d2d6c1f636a28b1e8f4e76cf92e43d1"}
{"id": "train_22233", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "reproducibility", "evaluation_metrics", "auditability"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e0c47d45d15ec40775403edf6e899b088aab687e"}
{"id": "train_22234", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "governance", "evaluation_metrics", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a8dc816b5aa8441ce32b29467f3393a80b981faf"}
{"id": "train_22235", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "auditability", "ci_integration", "tests_are_truth"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "11244905a78b10f8fd726ab60392971aef2b7b21"}
{"id": "train_22236", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "repo_scale_reasoning", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b04a0c1b0645a5d4222072492643561858f106cd"}
{"id": "train_22237", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["documentation", "ci_integration", "repo_scale_reasoning", "security_gates"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7b74d3ae7ef7080381dd85cbbeaf13eb1ea1de3e"}
{"id": "train_22238", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "tooling", "documentation", "security_gates"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "43f8736406b2352a7d507f9ed14af5d7dd34bc01"}
{"id": "train_22239", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "tests_are_truth", "security_gates", "governance"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d06f3495ee7302eab995396270149d3f2b7d6bcd"}
{"id": "train_22240", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "auditability", "governance"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a3a5bb61ef15fc0ddfcc8034727bd4297de7c4cc"}
{"id": "train_22241", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "repo_scale_reasoning", "documentation", "tests_are_truth"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8fe5546d156fe7b7de6cf545be786054e002baf1"}
{"id": "train_22242", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "evaluation_metrics", "governance", "documentation"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "37ba3767890ffb9c6c5409240ee56fb212e38f2f"}
{"id": "train_22243", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "documentation", "reproducibility", "security_gates"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3d5fbecca403f50f39a6305b1e125046878a88cc"}
{"id": "train_22244", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "repo_scale_reasoning", "tooling", "auditability"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d7fd4f15b5ad27ab4abd5c9708cf028e160a7d06"}
{"id": "train_22245", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "governance", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2ab5167126159f012fcbf45c441fb6ac1b24ffe8"}
{"id": "train_22246", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "auditability", "governance"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "309ad33a443775214001c85b41571fd4248e06b0"}
{"id": "train_22247", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "auditability", "evaluation_metrics", "security_gates"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "04bac78ba5f49f79fb1537faacc2e1d656feebd6"}
{"id": "train_22248", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "ci_integration", "security_gates", "tooling"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "24277e7f3f05a14aa44ba3eb077c1b2e47470313"}
{"id": "train_22249", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "ci_integration", "security_gates", "auditability"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4d25a3c21c865391fe7f4ae1de28c11a4b8d4458"}
{"id": "train_22250", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "evaluation_metrics", "documentation", "governance"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5c57ad020bba84876ad15164e04b58aa0220b66a"}
{"id": "train_22251", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "auditability", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0c626d9a506c1d8d3fedf8546f5b26a9d769243c"}
{"id": "train_22252", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "tooling", "tests_are_truth", "security_gates"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f2b185780732c289f517cb3553fbe11f7aa7f5ca"}
{"id": "train_22253", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "ci_integration", "tests_are_truth", "governance"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dc9b6219104e3a4ac12c8bb796663c2b09533ede"}
{"id": "train_22254", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "auditability", "repo_scale_reasoning", "documentation"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0ee20d4ab15c0f0f77b3f9d924e969745891661d"}
{"id": "train_22255", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "governance", "tooling", "documentation"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8d675c2cba3b597c9b9880d59acede0ab783a469"}
{"id": "train_22256", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "tooling", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8cf954c3ee446f90fd80caadf5cb8c9d2f37b32a"}
{"id": "train_22257", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "security_gates", "evaluation_metrics", "ci_integration"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e666cc1dafe2d9fea82d0b0c799109ee12d67c15"}
{"id": "train_22258", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "tests_are_truth", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5b9f7253288f843a14f66e9cda8fe154db898b43"}
{"id": "train_22259", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "documentation", "security_gates", "tooling"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "48d33959b0e7d343329e3834f63fe1fbbd9211e4"}
{"id": "train_22260", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "ci_integration", "documentation", "security_gates"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1de1bfb48e1dc6305d0b157d3d74996e35fe220c"}
{"id": "train_22261", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "reproducibility", "ci_integration", "auditability"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6015c701cab7f1a44fcc0d02d2641293830c3f6a"}
{"id": "train_22262", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["governance", "repo_scale_reasoning", "evaluation_metrics", "reproducibility"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ce894ea87822e90371a281ab37b53682c874b44e"}
{"id": "train_22263", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "evaluation_metrics", "security_gates"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b565c64d0c4a70d116134a166ed3ca30a9a709ee"}
{"id": "train_22264", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "documentation", "tooling"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7244ef29260d2dbb7f876dcf83a44e5b593cef55"}
{"id": "train_22265", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "ci_integration", "auditability", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e336a3603319105f01b5f8322b8539eaa47f4eb6"}
{"id": "train_22266", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "evaluation_metrics", "tooling", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "98dfce3a6aec6a60fa9f54d8e900c1f60f40b618"}
{"id": "train_22267", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "tests_are_truth", "reproducibility", "auditability"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c36bf0b2bae83f3b41b0cb5d1f7e15436e180964"}
{"id": "train_22268", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "tooling", "security_gates"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2f684fa80f91d2e8d5463370fc187d9d9fb8b7a4"}
{"id": "train_22269", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "governance", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1ee0315632a8ae9d6d952344b86403d758fc0dbd"}
{"id": "train_22270", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "reproducibility", "tooling", "documentation"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6b4a04e11ab88952b19de02c005f417cfb5044f2"}
{"id": "train_22271", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "evaluation_metrics", "ci_integration", "security_gates"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8681fc1d7512a29887208e4c6c6930a65327afe0"}
{"id": "train_22272", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "reproducibility", "cost_latency_tradeoffs", "governance"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4ac14f013f59755f329996b10f1159a6782d35ce"}
{"id": "train_22273", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "reproducibility", "security_gates", "governance"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "96cdc97daad920a59cf2792392e04aa8c3c8d4a6"}
{"id": "train_22274", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "ci_integration", "documentation", "tooling"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "eab970f17063ea791dfd9529a7c1c1d9f5efd8ba"}
{"id": "train_22275", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "documentation", "repo_scale_reasoning", "governance"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f90ae80819899e454b62971355359d56a1b472e7"}
{"id": "train_22276", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "auditability", "documentation", "ci_integration"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5abaa141de42746539be727019d41ec493bf591f"}
{"id": "train_22277", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "governance", "evaluation_metrics", "security_gates"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f7603927e5a2ffe1c7fa3c65abc2b76f04acfc50"}
{"id": "train_22278", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "cost_latency_tradeoffs", "evaluation_metrics", "repo_scale_reasoning"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cdc3032216d4a5ba022260ecd5075a3621586ec7"}
{"id": "train_22279", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "governance", "reproducibility"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1088129968bf7cb3a3cf1e051030e0a50832247b"}
{"id": "train_22280", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "documentation", "auditability", "tooling"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d8e12f098edf95ad8140b0f9afec75855b624ee7"}
{"id": "train_22281", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "ci_integration", "repo_scale_reasoning", "auditability"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ac2a7c1cc3fd180706a8ae7e98ff243d1daebe24"}
{"id": "train_22282", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "cost_latency_tradeoffs", "governance", "security_gates"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3cb663132613624453fb5c5ac57199b9f9ddb292"}
{"id": "train_22283", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "documentation", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c96801179aec92efc1ba3b109e18aedc4afd1261"}
{"id": "train_22284", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "security_gates", "evaluation_metrics", "ci_integration"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d7bc685197cc1ec7192d08293d60ad2a95d1007f"}
{"id": "train_22285", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "documentation", "security_gates", "evaluation_metrics"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f91e3472ea5f7dc6f00cef0f239d8f08b04db2d3"}
{"id": "train_22286", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "auditability", "security_gates", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a5acb3ed40a7131e9897c5b65ffc6937bdaa8e23"}
{"id": "train_22287", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "evaluation_metrics", "documentation", "tooling"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6b7ac913b5f1ccd2e32a7cfcbf7b8999c7f91c40"}
{"id": "train_22288", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "security_gates", "evaluation_metrics", "auditability"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bc4a111abcf7ee48b7b2fd7632c20f95bfe8abb1"}
{"id": "train_22289", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["governance", "documentation", "ci_integration", "tests_are_truth"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "67739a46ce02aa93610412f5242d52a2053eb0ed"}
{"id": "train_22290", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "governance", "evaluation_metrics", "security_gates"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "964d7c9f360dde82d78ca2a2868ca3847cbd65cb"}
{"id": "train_22291", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "auditability", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d5a42477005db143bc9b083c1319ed53ebef360e"}
{"id": "train_22292", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "tooling", "auditability", "governance"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "81f67d5f3ce3277ff1233f341d7b9aae94d37ba4"}
{"id": "train_22293", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "cost_latency_tradeoffs", "reproducibility", "security_gates"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bcf60f41bb45adaeb677cf4d4e28a80ca86b823b"}
{"id": "train_22294", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "evaluation_metrics", "ci_integration", "documentation"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a624f34d119a4ce9c289f708aeac91250e7c36c0"}
{"id": "train_22295", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "governance", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3047e1260535ade3e15ba60ab792574b1a9ebe2a"}
{"id": "train_22296", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "ci_integration", "security_gates", "documentation"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "931322d5c22cdbdb5db412d784b6317a942fec89"}
{"id": "train_22297", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "tooling", "reproducibility", "auditability"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "afe9d66f55feb6c1dbd4ca62199155dc07be4d40"}
{"id": "train_22298", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "auditability", "documentation", "reproducibility"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f1e12e80330b25794caedfcee08452fb87d50507"}
{"id": "train_22299", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "reproducibility", "repo_scale_reasoning", "security_gates"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "987516d6c148eb6acd120e005f5f6b9f2615f609"}
{"id": "train_22300", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "governance", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fd45ef5f8e3c62b39b203b938eb00c4c667e77c8"}
{"id": "train_22301", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "tooling", "security_gates"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1ced8645d458f12c1cbdd27f87896fc9ff1da887"}
{"id": "train_22302", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "ci_integration", "repo_scale_reasoning", "reproducibility"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9decf9247421cd9358c613d92e44f3a48e75ac5a"}
{"id": "train_22303", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "auditability", "evaluation_metrics", "documentation"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dfd88830af0bb79d760979775124360fa51ef7a7"}
{"id": "train_22304", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "ci_integration", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "181654a103c17503eebf871c648637e5458feb7f"}
{"id": "train_22305", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "governance", "ci_integration", "documentation"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a59c6c72eb00fa0b7f2b9b37c404668e41638ffe"}
{"id": "train_22306", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "repo_scale_reasoning", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "51fe12ec33ba690b357503d4064392ce97e11145"}
{"id": "train_22307", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "repo_scale_reasoning", "governance", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3c275adc0c5e6747b88dad1e41bd60f22b01560a"}
{"id": "train_22308", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "tests_are_truth", "security_gates", "ci_integration"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "24b226461f1a99a09fad24691f95a3870668d2b1"}
{"id": "train_22309", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "cost_latency_tradeoffs", "ci_integration", "tooling"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "82061c3b512ad5604af7185008ed8b85cfe7e12e"}
{"id": "train_22310", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "cost_latency_tradeoffs", "documentation", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2ac2b4a05447038b3e4b2095112735ba8d14c986"}
{"id": "train_22311", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "evaluation_metrics", "tooling", "security_gates"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "22dcebd45fcaee5310662bcdf5ed9d0927265df6"}
{"id": "train_22312", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "evaluation_metrics", "ci_integration", "auditability"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c64bb68a97dc5c53d9f44faf55ee1773a9099c3e"}
{"id": "train_22313", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "auditability", "documentation", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "212217a3675017849fdb6c8737ff132fdee81c23"}
{"id": "train_22314", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "reproducibility", "tooling", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "59e23404751c9149fcc8af5cf103ee695e8ab37b"}
{"id": "train_22315", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "security_gates", "tests_are_truth", "tooling"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bdad78149e4ebff93301b9a891b9d14bc85e623e"}
{"id": "train_22316", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["governance", "reproducibility", "evaluation_metrics", "tests_are_truth"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "06df810d55d985bca61760a2fd4b6482cd71e92e"}
{"id": "train_22317", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "documentation", "tooling", "governance"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1e55daf54fc321714f54cc0b48620468315c325f"}
{"id": "train_22318", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "reproducibility", "ci_integration", "governance"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4e712233f7ea7299114cae93731bbb436b9d8b5b"}
{"id": "train_22319", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "tooling", "repo_scale_reasoning", "reproducibility"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3227e04e2bcbaa7e8ea07e7316faf4c57d63e4f6"}
{"id": "train_22320", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "governance", "evaluation_metrics"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e0a7cd9730e8fc0039ea8f495799dbf3cfeb519a"}
{"id": "train_22321", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "reproducibility", "governance"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "491bb12ceca6d9c181efcfff96881f28748bafc1"}
{"id": "train_22322", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "evaluation_metrics", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7b6f4e1cff995c66b03f523866b5b2b5ef4226a3"}
{"id": "train_22323", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "ci_integration", "evaluation_metrics", "security_gates"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2bc02e926b848cd351036f311657cd88231b391d"}
{"id": "train_22324", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "tests_are_truth", "tooling", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4312f8c9026018b9aca6e3f16a5dbdb5b5d9f351"}
{"id": "train_22325", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "ci_integration", "security_gates"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0386cccae27f17e584f60e1a27ad94fddf7ad7fc"}
{"id": "train_22326", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "tests_are_truth", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "57abcb1194ef9fc8014bc9b8c30a203b466de7fd"}
{"id": "train_22327", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "documentation", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7bad926e6a988ec1989932cb6bccc8f76c331742"}
{"id": "train_22328", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "governance", "cost_latency_tradeoffs", "security_gates"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d42f8993f8a9471815a6cf11dc9a65c852450cc8"}
{"id": "train_22329", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "cost_latency_tradeoffs", "governance", "repo_scale_reasoning"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "982fb8383ca591d7e7fc2e89f8bc554eb5191eb1"}
{"id": "train_22330", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "security_gates", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "28f17f9bcacba5eb8dc2cff27e675224dbdd8c0e"}
{"id": "train_22331", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "evaluation_metrics", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d87e46793fe013655c00001299bab1b929455a64"}
{"id": "train_22332", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["governance", "reproducibility", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "95e6cf5a69b45016cf703ee9145ee6644f9b6b95"}
{"id": "train_22333", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "tooling", "documentation", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1e4fb9ad7499197706424189cdc0a43435fbf876"}
{"id": "train_22334", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "documentation", "tests_are_truth", "ci_integration"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e61824c4b98f4518e9bd913b4a09ce3883f0efad"}
{"id": "train_22335", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "tests_are_truth", "repo_scale_reasoning", "reproducibility"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2cd07960103df59b4fbf0992e825c1ee986d5fa4"}
{"id": "train_22336", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "tests_are_truth", "ci_integration", "governance"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3cf7bcf5ac1d9983f76b892b922601940de7a97a"}
{"id": "train_22337", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "security_gates", "repo_scale_reasoning", "reproducibility"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dfcb2610e7d8e104a59e146592af25715189e507"}
{"id": "train_22338", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "tooling", "reproducibility"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2a9e2f2a82745b26a2c0071c0853d9be007a5c87"}
{"id": "train_22339", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "governance", "tests_are_truth", "tooling"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2564163c542d34646faaa2fe4e90e539ea799399"}
{"id": "train_22340", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["documentation", "ci_integration", "governance", "tooling"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1acf6b40ecf40315cba9f1402118ad4f26388b76"}
{"id": "train_22341", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["auditability", "governance", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "272693e8fb9d45ed7e827a06d1b8663327a0b22b"}
{"id": "train_22342", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "evaluation_metrics", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ca7afc13784cc985cf0e061ed78d821e80efb80b"}
{"id": "train_22343", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "governance", "documentation", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ab56a03f4e2200672e8008d735a478e2c438d68a"}
{"id": "train_22344", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "security_gates", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "115eed8757156b2a2161ced46c8a18fe247f4b5f"}
{"id": "train_22345", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "security_gates", "auditability", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "55508163eb1d094c2a9b76fb3a4fc99648beba3c"}
{"id": "train_22346", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "tests_are_truth", "tooling", "security_gates"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e339e4d58ad21be824abb49000a294e707201fb8"}
{"id": "train_22347", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "tooling", "governance", "ci_integration"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b6fe28c6b240180a69666b904d31812548e52a0b"}
{"id": "train_22348", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "repo_scale_reasoning", "auditability", "security_gates"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0fc5be51de9b6a82a541a54d4659772d4ebd7966"}
{"id": "train_22349", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["auditability", "tooling", "evaluation_metrics", "documentation"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4523bab3d6f137c7681a1b8ff1cb00eddfca8878"}
{"id": "train_22350", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "security_gates", "tooling", "documentation"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8f9c095c419df31593a30943e677218b49119823"}
{"id": "train_22351", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "governance", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "14666bbecc61fb1426a36ef6e7c15acd36b446c9"}
{"id": "train_22352", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "tooling", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2c1fb858fcd83621123ecaf5098a908901eda622"}
{"id": "train_22353", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "repo_scale_reasoning", "tests_are_truth", "ci_integration"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8636f453bf2c6a18cac5086e812318118df1a108"}
{"id": "train_22354", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "auditability", "documentation", "ci_integration"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "50e6a711ea170b192a3075011efbcbfd0ab2ebaf"}
{"id": "train_22355", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "reproducibility", "repo_scale_reasoning", "governance"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a2f8bb7fa6a2516432f8afa285efef1cd6e6c322"}
{"id": "train_22356", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "ci_integration", "repo_scale_reasoning", "governance"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "03ba8e448ae8db4c92e6e38bc9a9b21ba83c8244"}
{"id": "train_22357", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "documentation", "tooling", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "60d2b4438119fde49252621785258e271f72c236"}
{"id": "train_22358", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "security_gates", "evaluation_metrics", "documentation"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "94d79483b69c2fd7b6c2c812d829a095e9a74ff0"}
{"id": "train_22359", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "governance", "repo_scale_reasoning", "auditability"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d16ac23450bcde062ef293d89e1e6de7c3755676"}
{"id": "train_22360", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["governance", "repo_scale_reasoning", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ef589e6a9d74704efe6bb12bb6a22c9b1d7cd7a8"}
{"id": "train_22361", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "auditability", "reproducibility"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ae27cb0d0f140772f825fb3c78acae4f5f8cf1ab"}
{"id": "train_22362", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "tooling", "documentation", "governance"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a6a501cde4451f9e4f8de0f9a478299776547336"}
{"id": "train_22363", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "ci_integration", "reproducibility", "tooling"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6cd8cc0b90b366a96c31186d62f7a12b834e1bee"}
{"id": "train_22364", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "reproducibility", "documentation", "evaluation_metrics"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "757af2b70b42c21102197b66ce1f2948c5913701"}
{"id": "train_22365", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "evaluation_metrics", "security_gates", "auditability"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e18f2807c09e8c3139075a9d8f7b50521ac772bc"}
{"id": "train_22366", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "security_gates", "cost_latency_tradeoffs", "governance"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0587f3ac5dac187b5a1653a6a0b222578c7f5492"}
{"id": "train_22367", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "repo_scale_reasoning", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "518c73c3e1f23c2a9df99e7dc2924cc2eedf25bf"}
{"id": "train_22368", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "tests_are_truth", "reproducibility", "governance"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5c550ba6e26c1ae27c14fbb2d41c76d984851aff"}
{"id": "train_22369", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "ci_integration", "evaluation_metrics", "governance"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "eae855e82252b0268fedf0b80b0a4f3e70dae5f2"}
{"id": "train_22370", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "security_gates", "reproducibility", "auditability"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "02d8f635deb4441a222700f65a7ec45e3b233d3f"}
{"id": "train_22371", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "security_gates", "ci_integration", "governance"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "87ffd06b18fe05d88f777f68b9dafb65659945cc"}
{"id": "train_22372", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "auditability", "tooling", "governance"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3824e6b3c3f1bf750f577a3d6a73fb6950016eb8"}
{"id": "train_22373", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "governance", "documentation", "security_gates"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3b41522eed59c4d83111f0da4d152b3a3e1f6f7c"}
{"id": "train_22374", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "cost_latency_tradeoffs", "repo_scale_reasoning", "security_gates"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "73d06da98b55a7502cfa0098e28e446de25c7e18"}
{"id": "train_22375", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "governance", "tooling", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "87e291ad9394b7d1c91ce6895845f06deafb7c21"}
{"id": "train_22376", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "tooling", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c9b97358edd69c5b1a1a455418dda5af4811f07e"}
{"id": "train_22377", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "repo_scale_reasoning", "reproducibility", "evaluation_metrics"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f61e2a2a02de9a92333c23196ad0ea7499150c7c"}
{"id": "train_22378", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "auditability", "ci_integration", "documentation"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "191e1fe0681a5a372407c3bb965b98386872c2ee"}
{"id": "train_22379", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "auditability", "security_gates", "governance"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d73c41fa9196d589999cef076b286c89f61c3366"}
{"id": "train_22380", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "governance", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "454fc72d206887ba8d3aa95e5345dc6d6385c5c9"}
{"id": "train_22381", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "auditability", "tooling", "security_gates"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "85683775fa7c2b18c0dfd27a3148cb8cf591b3ad"}
{"id": "train_22382", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "auditability", "documentation"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0f4a1dde249dc2711a5b6fe74b617eb36662ce08"}
{"id": "train_22383", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "auditability", "security_gates"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "eedd1f39d6b019cbc31e9dacac7f1d552b1470b4"}
{"id": "train_22384", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "ci_integration", "governance", "evaluation_metrics"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "383a5926a419c8e074c089859a2447f6021febbd"}
{"id": "train_22385", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "evaluation_metrics", "governance", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "727dd238e03a35d7233194c480eefdc10bf317c3"}
{"id": "train_22386", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "ci_integration", "documentation", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f856e4bc5a75a2c32a7dd000ef986a08e6ddacaa"}
{"id": "train_22387", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "security_gates", "governance", "documentation"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9af7f6b1dafcd90ee034d9272a915f33ffe2e885"}
{"id": "train_22388", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "tests_are_truth", "reproducibility"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8dae5f6bb99517deba42f6a88f668b02c19166bf"}
{"id": "train_22389", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "reproducibility", "auditability", "documentation"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0605fbc7b313934d6261e15211f12d063afc70b3"}
{"id": "train_22390", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "evaluation_metrics", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1455df681e206a8e5dc5bc9e752d2f03b48f0c84"}
{"id": "train_22391", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["governance", "documentation", "tooling", "tests_are_truth"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a5859e6a49df1135e92573ab00d3e480f5cec1a8"}
{"id": "train_22392", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "tooling", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "eb307ff479ba6e376126f4e619ecd2a6877e7f42"}
{"id": "train_22393", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "documentation", "repo_scale_reasoning", "tooling"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7fa36afeb62e9deddd8cccf36ba0def7f317576c"}
{"id": "train_22394", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "auditability", "security_gates", "tooling"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e2003591781cf3aa13cc8b7958ea7292b8c65ced"}
{"id": "train_22395", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "ci_integration", "reproducibility", "security_gates"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "324b40e4020a412b2da0d8731b0b107977d302fb"}
{"id": "train_22396", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "documentation", "ci_integration", "security_gates"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "497ed367a3dab335b8ccfbe5dedaf111d78902d9"}
{"id": "train_22397", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "reproducibility", "auditability"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "564a8e0e35843a23efc81d46a56a3f18ed71424c"}
{"id": "train_22398", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "repo_scale_reasoning", "documentation", "tooling"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b8c346b5bbd029a4d526e0b5d5150bc88b637a1b"}
{"id": "train_22399", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "auditability", "tests_are_truth", "documentation"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3b91111cbbdd4de8ae201572ac3f99be20a3d118"}
{"id": "train_22400", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "repo_scale_reasoning", "auditability", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "729748278121ca6c1928a4a1ee5032aae11da183"}
{"id": "train_22401", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "evaluation_metrics", "auditability", "reproducibility"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "92a37f59bbaa9cffee440d96c6712ce44a2768e0"}
{"id": "train_22402", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "governance", "evaluation_metrics", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ae8085f98813a15fffb3635331673b17a36641c5"}
{"id": "train_22403", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["auditability", "governance", "security_gates", "documentation"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "128c2e2d82f786a8ed904c413595f7a56e5bf5b1"}
{"id": "train_22404", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "security_gates", "evaluation_metrics", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a58e065cee175022e7bf9c4a9510235c517cd881"}
{"id": "train_22405", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "evaluation_metrics", "governance", "tooling"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7ea34a72c81cab07160542f0b862224dc5113cfa"}
{"id": "train_22406", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "documentation", "tooling", "repo_scale_reasoning"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "67b4839cd5c6e744a9d3dd02668e36dff96955f4"}
{"id": "train_22407", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "security_gates", "auditability", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "68f43362a197570d34b61828b5aca64380d3c935"}
{"id": "train_22408", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "cost_latency_tradeoffs", "evaluation_metrics", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f8ee780588a5c922213cbfcba72a70eaf0cb1516"}
{"id": "train_22409", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "tooling", "governance", "tests_are_truth"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5367ad6b3eb2718bc5e4c293238f97e00c4f79bc"}
{"id": "train_22410", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "documentation", "security_gates", "auditability"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3d2ab922dbdec1786b81fe5a407b9ed12b0b4151"}
{"id": "train_22411", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "governance", "ci_integration", "security_gates"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "50f449d6f9eb41588c6afeca595d36e98735a6ee"}
{"id": "train_22412", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "governance", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "aed77474065264ea1810cb9ac5077952863d544a"}
{"id": "train_22413", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "reproducibility", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d6de4bb0c2e8dfca04fd098de26fb6c28ab19b29"}
{"id": "train_22414", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "auditability", "repo_scale_reasoning", "governance"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "beaf1d40a37698ad1a4f47e7fc7a5179f75df283"}
{"id": "train_22415", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "auditability", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "40693deb3c0ac8ca0d9b6e157863a3e1294270e7"}
{"id": "train_22416", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "ci_integration", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dfe8ebe93e99b5be59e921ca1eb349f9796fdfbf"}
{"id": "train_22417", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "tooling", "reproducibility", "evaluation_metrics"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a3712a8f19f2479cafb23d9abe86eea321ff9620"}
{"id": "train_22418", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "governance", "documentation", "tooling"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ae8085f98813a15fffb3635331673b17a36641c5"}
{"id": "train_22419", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "security_gates", "reproducibility", "auditability"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c3a8fab7bca6eeaa6a729c709471146e7592a29b"}
{"id": "train_22420", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "documentation", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "78a240c619e759d8601f264b06d47066b46bcb3c"}
{"id": "train_22421", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "security_gates", "governance"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c74a23897e86583c0a127778f45dd501330dcdaf"}
{"id": "train_22422", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "tooling", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1bcc9a385ded9b4499237ae72296f8b25ec048c5"}
{"id": "train_22423", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "auditability", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "92124498431062885fab14b148061ec6105b3677"}
{"id": "train_22424", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "repo_scale_reasoning", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "970431286f0083848a424eb3dbb2d80c1cf870d6"}
{"id": "train_22425", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "governance", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9092493249bf5ad15d4efbc2bba2982a55c54031"}
{"id": "train_22426", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "reproducibility", "documentation", "tests_are_truth"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "43d235f0578dcc347e1a9a73ae6d3e89d9e1cf2d"}
{"id": "train_22427", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "reproducibility", "ci_integration", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9d57c87800b1f74ca085e5c485d2b450d5a609e6"}
{"id": "train_22428", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "governance", "documentation", "security_gates"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dede9eba7d2283e8eb936d62b3fb96733cf8a97b"}
{"id": "train_22429", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "auditability", "governance", "tests_are_truth"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cd9b70c2f7da16c57c5fac38994d27238bc6912c"}
{"id": "train_22430", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "reproducibility", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ad22dbbd04fdbfdeb6ac521cb9c3a35075c81a7d"}
{"id": "train_22431", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "auditability", "governance"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a25917c9a43b9a2767428c0600a14bd3c57f2777"}
{"id": "train_22432", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "governance", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f06365e4053e479b9e7247943ea738043b43ea62"}
{"id": "train_22433", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "governance", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ddcd471e6318b2fcdd152fc7c0b5d8ee84e3ba01"}
{"id": "train_22434", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "tooling", "auditability", "documentation"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "74632c382fac51a0fdb9c771febb3504b55e2faf"}
{"id": "train_22435", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "auditability", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "56e1a672b6dfac59e8156153177d9cc70bd546ae"}
{"id": "train_22436", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "evaluation_metrics", "governance", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1bcddf25419b5070c134cb329fda83c006f9ef84"}
{"id": "train_22437", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "governance", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "34626fa2bb670662f92380e852e909a0a264df78"}
{"id": "train_22438", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "976d8267fe20e1c794c71df75b14cd4491fec68f"}
{"id": "train_22439", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "governance", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e523bd2a9474230c0e30815f4210eb0d9709c656"}
{"id": "train_22440", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "tests_are_truth", "cost_latency_tradeoffs", "governance"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "176f39da7a808b3be7d1687bddb25ece5672b979"}
{"id": "train_22441", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "tooling", "governance", "documentation"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9bb915b0610119ae27845b79bf266482a222de95"}
{"id": "train_22442", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "tooling", "security_gates", "reproducibility"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2e2ed6cb2f0554868c148743af3783c619729d20"}
{"id": "train_22443", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "governance", "documentation", "ci_integration"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "70f7b4e72b3e7b439acdaf00bd7b90b5dc989d3e"}
{"id": "train_22444", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "repo_scale_reasoning", "documentation", "reproducibility"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0d3535cc82606ba98c337c5eeae0f160b930d61e"}
{"id": "train_22445", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "tests_are_truth", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1ae78269b8c7551fb70cac8c547462cdd6d9efca"}
{"id": "train_22446", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "documentation", "repo_scale_reasoning", "reproducibility"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e6a0f5788d306ba692ecccabab1860adf68b8ff8"}
{"id": "train_22447", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b4a377a20f62c86d569dc326d22e4871c8dfd3c5"}
{"id": "train_22448", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "tests_are_truth", "governance"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ab31c6fa2c699ad8a7c54b07f5892fb231c73981"}
{"id": "train_22449", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "ci_integration", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "529ef92d8d8b12f2347b70e48ec1447c53b2e9ed"}
{"id": "train_22450", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "repo_scale_reasoning", "governance", "ci_integration"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2052b0f1f3225f1bd669e55ab96599f868c2aca1"}
{"id": "train_22451", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "auditability", "documentation", "repo_scale_reasoning"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9d57c87800b1f74ca085e5c485d2b450d5a609e6"}
{"id": "train_22452", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["governance", "tests_are_truth", "documentation", "security_gates"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6fd3e14186ec3578a9731ce0f161baad265342b7"}
{"id": "train_22453", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "security_gates", "tests_are_truth", "ci_integration"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2bf33268f78ba19c54e7afe07c561af60e7f3628"}
{"id": "train_22454", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "cost_latency_tradeoffs", "ci_integration", "reproducibility"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a870d79cb12c5888e0f3c43f8e6f057caa9b683a"}
{"id": "train_22455", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "tests_are_truth", "governance", "security_gates"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0704659571c1be941dd269ab1cc9e35e1eeb51b9"}
{"id": "train_22456", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "repo_scale_reasoning", "documentation", "auditability"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c0e216080364cab431a47b77a28bc73f577da022"}
{"id": "train_22457", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "tests_are_truth", "auditability", "security_gates"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f5ca31cd3bc43de20aaff70be64426c970d7a8d1"}
{"id": "train_22458", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "repo_scale_reasoning", "reproducibility"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7bbb9f54329a0d9eda29395a850ee6d052b13edb"}
{"id": "train_22459", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "evaluation_metrics", "auditability", "documentation"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5711bc29ac6c15ed4f45dfc8da30cdd7c431aea5"}
{"id": "train_22460", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "security_gates", "evaluation_metrics", "governance"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b1ee9d746940f7152b55cb6b138d538808e2b7a3"}
{"id": "train_22461", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "ci_integration", "tooling", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7509a4795ad57bee202c645de78f05e1cb790b01"}
{"id": "train_22462", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "tooling", "repo_scale_reasoning", "documentation"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0d3f491449d7093d4bbb3a8abe4fe59b4b2fc945"}
{"id": "train_22463", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "auditability", "governance", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a2385e9a737650245d77a2ad0104c9846bf0e8c6"}
{"id": "train_22464", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "tests_are_truth", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "04fac14212ae1f07e1c241f7c2615f11f1fd1efd"}
{"id": "train_22465", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["governance", "tooling", "security_gates", "reproducibility"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bd44d90360a3ce318d228b25e28000df4917f49a"}
{"id": "train_22466", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "security_gates", "documentation", "auditability"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "20330911737e2557a72a2e0deec84bd59c209501"}
{"id": "train_22467", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "tooling", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c9f65ddbccd1f41c5ac832ef781358c8f0bc5a52"}
{"id": "train_22468", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "tests_are_truth", "evaluation_metrics", "ci_integration"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2fdba594cf6f0f0c585f2171f494a76b422890be"}
{"id": "train_22469", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "security_gates", "tests_are_truth"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "91b8f0d135ea3986fbd91ab44ffe1e26ef82a0b8"}
{"id": "train_22470", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8cd631a326acb77a94738a1dd99ad973664e0a6a"}
{"id": "train_22471", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "governance", "ci_integration", "auditability"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cc1f6dce182bdbda2994192940f591b0ed5b529b"}
{"id": "train_22472", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "evaluation_metrics", "security_gates"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1852f869dea93df0a0e64a59277212bdbf01dd97"}
{"id": "train_22473", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "tests_are_truth", "documentation", "tooling"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b49eadb2b2c4f8b7a458d12e7f7f214683fef6b0"}
{"id": "train_22474", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "security_gates", "ci_integration", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ed6367eae566f508b3e0adf10442c710f55f4535"}
{"id": "train_22475", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "evaluation_metrics", "security_gates", "tooling"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6db968525e672993fbc6c3770eeed66b8e4bdcc9"}
{"id": "train_22476", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "ci_integration", "evaluation_metrics", "documentation"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5ac2fdf94c7ad304df174bf565f9cc9dafad4fa7"}
{"id": "train_22477", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "repo_scale_reasoning", "tests_are_truth", "reproducibility"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "09ee32c8858101348849d4514982b8e919dc2ec1"}
{"id": "train_22478", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "ci_integration", "documentation"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "39f29b3d1fe9697d4e78dbeaefbf457878b4c023"}
{"id": "train_22479", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "auditability", "security_gates", "tooling"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d9665f28a65d85a551b46cfbf47cc59f7340ad64"}
{"id": "train_22480", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "documentation", "tooling", "security_gates"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7bd75bae6578c550d1dfd76d93ef04032633f919"}
{"id": "train_22481", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "documentation", "auditability", "reproducibility"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a8618fb673bcfe08923ffcaad4612b9eb2a04008"}
{"id": "train_22482", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "ci_integration", "governance", "tests_are_truth"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "db2d551cd80957ff0c52f9638e63a6ed3618a381"}
{"id": "train_22483", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "evaluation_metrics", "tests_are_truth", "documentation"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "78dd12677defa01d706b5affbfc7b810912f68a3"}
{"id": "train_22484", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "reproducibility", "cost_latency_tradeoffs", "governance"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "89dc102b5de1777863106856bf3753d8502d27e3"}
{"id": "train_22485", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "security_gates", "governance"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bb61456e5266a7d285539fe9d2196c690c142a18"}
{"id": "train_22486", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "security_gates", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "09b705f8b75dd5d411c21394691d79e230692c86"}
{"id": "train_22487", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "documentation", "reproducibility", "governance"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "654f847fd62e18188e00b0abfd2670821808b034"}
{"id": "train_22488", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "security_gates", "reproducibility"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "731ed6c76645625e11aeed0cfb8fbc0e52026d5f"}
{"id": "train_22489", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "reproducibility", "evaluation_metrics", "tests_are_truth"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d35c085e93f49917ab196296906ee21dc5b6ffe3"}
{"id": "train_22490", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "tests_are_truth", "security_gates", "tooling"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fe12cc3492f49edb2ddc23dcbbcbb423c11d6230"}
{"id": "train_22491", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "reproducibility", "governance"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b33c215031445e791a4de43f8f85fd3988dcfd3d"}
{"id": "train_22492", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "governance", "tooling"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2364e9fe8c825937a938440e853ccac7c42fb2b6"}
{"id": "train_22493", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "ci_integration", "reproducibility", "evaluation_metrics"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2971397baac89d3d00cbc258453f08cee33f7afd"}
{"id": "train_22494", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "repo_scale_reasoning", "auditability", "ci_integration"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fb27f496e101c0e14cdb9884b65f95ca2e489060"}
{"id": "train_22495", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "tests_are_truth", "auditability", "documentation"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "88cd211d7cefa9322cf3727027eb1a13d3f8ef13"}
{"id": "train_22496", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "security_gates", "auditability", "reproducibility"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "73bbf05a5fbc2c4c1edc91bd1eacbbfd5e7f9d8a"}
{"id": "train_22497", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "tests_are_truth", "ci_integration"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0f74506087ee30f588e8a7b85dd00e742cc6c6f3"}
{"id": "train_22498", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "tooling", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c0e1753cf47df3bd13fbda5ec1cb4369e10bd754"}
{"id": "train_22499", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "reproducibility", "cost_latency_tradeoffs", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "77e472682229cbb37dd7f7c189d4e2a094d34dea"}
{"id": "train_22500", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "ci_integration", "reproducibility", "governance"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c3ff5b71975621607ae4e19ecfd5691ed890feed"}
{"id": "train_22501", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "auditability", "tooling", "ci_integration"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3e2713b3cd5586f1778481fa9044da6da980fd80"}
{"id": "train_22502", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "tooling", "ci_integration", "tests_are_truth"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "df1d6a720dabf438611ce1cabf45ab4dea089958"}
{"id": "train_22503", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "ci_integration", "evaluation_metrics", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "65b60ebd808ca65d492450c645e77d0b0946a080"}
{"id": "train_22504", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "governance", "documentation"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2ac6908ebb5cadd80739a1c94ae053c7037ef58d"}
{"id": "train_22505", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "cost_latency_tradeoffs", "security_gates", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c6d9424ea8651ac0b7523d627c4c45d0f006dbaf"}
{"id": "train_22506", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "auditability", "evaluation_metrics", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "884de04cbe6cd9adff950e0cd38afcd5cfa24157"}
{"id": "train_22507", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "documentation", "governance", "ci_integration"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a593164f1d58d2ff51f4b2434190f20868255a66"}
{"id": "train_22508", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "reproducibility", "governance", "security_gates"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "25a2c563bb0fc6cd80520e15b6f342fdce8334c3"}
{"id": "train_22509", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "documentation", "tests_are_truth", "auditability"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5fa965e763290722a84996a5f42422a35c08c3ae"}
{"id": "train_22510", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "tooling", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fde0599797bace6e4918f207f56a4ee43ff80c6f"}
{"id": "train_22511", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "evaluation_metrics", "security_gates", "governance"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b75788aba9ea859cf9df25bca91594fa8bcb21e7"}
{"id": "train_22512", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "governance", "cost_latency_tradeoffs", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8c1dbffde1b5bb3dd3eb6b907b748eebb1617f70"}
{"id": "train_22513", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "governance", "reproducibility", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "da02b02c3befc88b8454e567ca2f11a8ea7e8505"}
{"id": "train_22514", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["auditability", "tooling", "evaluation_metrics", "ci_integration"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f2ceeadaa207f0c19e71e34144cbcf2066334d78"}
{"id": "train_22515", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "tests_are_truth", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "493bc07c6a1053420e551c8693b99a0bdd0d7f60"}
{"id": "train_22516", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "ci_integration", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b17c3b56a5e2079ce1f85bd94e0ecae9a30203e0"}
{"id": "train_22517", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "auditability", "tooling"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c68a9e5bfc00f2cfd4b383489d7ac3c12feaa4a2"}
{"id": "train_22518", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "auditability", "ci_integration", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ab32dacbbdcc3852d0156584e4fa3b52b37bdb4e"}
{"id": "train_22519", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "reproducibility", "evaluation_metrics", "ci_integration"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4c4eaec6de5d133cd6e756486b15bce9b8113ed8"}
{"id": "train_22520", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "governance", "ci_integration"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "304db5a7c62fe821b5d309dd872b4247ee70fc25"}
{"id": "train_22521", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["governance", "tooling", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "87e0363b144ae08a48476f3c549e0a84643d828d"}
{"id": "train_22522", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "tooling", "reproducibility", "tests_are_truth"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a8972a70041198cfde526c64f7fe5e03d6a0bfd2"}
{"id": "train_22523", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "reproducibility", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f7c920d4b9076f3cdc69283a456cde4b4daf51eb"}
{"id": "train_22524", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "auditability", "tests_are_truth", "documentation"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a122310c146114cabd6c37c5ef81f6c84b5ff0b3"}
{"id": "train_22525", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "evaluation_metrics", "security_gates", "tests_are_truth"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "21247bf6ec5b436158de9047059e8f59fb2c5d80"}
{"id": "train_22526", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "auditability", "documentation", "tooling"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6da05d5131a84307ea9846f23a51569129bc9639"}
{"id": "train_22527", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "auditability", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "844c41d393dad236bdc5872439f6729297f6f58c"}
{"id": "train_22528", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "auditability", "ci_integration", "reproducibility"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cc13d560a57145418916f5014c90631f2caee234"}
{"id": "train_22529", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "tooling", "auditability"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "843bcc86f6752dd1a8aa48cca36ebbeedab5c1c6"}
{"id": "train_22530", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "tooling", "repo_scale_reasoning", "reproducibility"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6c6d6836ef8d669f400998dec76ff4aa38989078"}
{"id": "train_22531", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "reproducibility", "documentation", "auditability"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dc9b36e8e8a573df7e585a4c1d27951eb47811bc"}
{"id": "train_22532", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "tooling", "governance", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "148e5b8fb42d47bd234d72d6afa18a261f2f6cf8"}
{"id": "train_22533", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "tooling", "reproducibility", "auditability"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f87b5c9836d99fcb8cc948b33966de0c18d56c13"}
{"id": "train_22534", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "tests_are_truth", "ci_integration", "governance"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "46628ac18c7bbadf99b31edbac3ce6781a96d4a2"}
{"id": "train_22535", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "security_gates", "reproducibility", "documentation"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8cb0a62093300a0a38ffa8dcc6fe96fb9f090ba2"}
{"id": "train_22536", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["documentation", "evaluation_metrics", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3b10853b34f9a73c1069e62f011997bdfd354d9e"}
{"id": "train_22537", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "evaluation_metrics", "ci_integration", "auditability"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6066658458db8fafc394a3afcbb79d27eb5014cf"}
{"id": "train_22538", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "governance", "evaluation_metrics", "tests_are_truth"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6d30326c1583af2bf5677e6ea8d79a119329ccac"}
{"id": "train_22539", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "reproducibility", "tests_are_truth", "ci_integration"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b7fa33e6021dc39bae6c189e2ca4e4ec7a1984a9"}
{"id": "train_22540", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "tests_are_truth", "tooling", "auditability"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "71d47e8ba2995e4b90a618c7a41287068e5618f0"}
{"id": "train_22541", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "ci_integration", "evaluation_metrics", "governance"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6b4b8dc4e94bcd9b09f60040e20c1ca6b7129644"}
{"id": "train_22542", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["governance", "cost_latency_tradeoffs", "reproducibility", "security_gates"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "afad5f6f483ae21944de9ab65904b836a5b0c357"}
{"id": "train_22543", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "documentation", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "340753b6d0d4c50877bb08dbea4bd92ed9531d1b"}
{"id": "train_22544", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "documentation", "reproducibility"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "60d1d530396e6904e18bd8b2a8ce5554fd89dd30"}
{"id": "train_22545", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "auditability", "ci_integration"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "022dc1f80255788a950a7dc202c457914449cb64"}
{"id": "train_22546", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "reproducibility", "documentation", "auditability"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "750886e5b3c16dcda42ed6086f7a2a7cc968ba69"}
{"id": "train_22547", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "tooling", "reproducibility", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "90c4171bf325ba0030a0a7c6003bf1aaea8c00c5"}
{"id": "train_22548", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "auditability", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2b1fa6fcf05720be52c2c8fdcd3bdc34f49ed982"}
{"id": "train_22549", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "reproducibility", "security_gates"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "420ff1a7e25c629c720dd2164672b4a83f46f958"}
{"id": "train_22550", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["governance", "auditability", "reproducibility", "ci_integration"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6abe9edb688c087a6ca9ab70a6deb761dcba7f32"}
{"id": "train_22551", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "security_gates", "tooling", "governance"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "996341ae10d5eb571a7b256df1b044ec3eb015cf"}
{"id": "train_22552", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "repo_scale_reasoning", "reproducibility", "tooling"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7c6856c5871bc8b9f46df5cf33fa1c26a142fcc0"}
{"id": "train_22553", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "auditability", "repo_scale_reasoning", "reproducibility"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e8750b697a845315552380f221b7ae1cb0b6bf63"}
{"id": "train_22554", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "evaluation_metrics", "tests_are_truth", "security_gates"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fe07213fdbf5a531c25bf2c6f2d732da21e391f4"}
{"id": "train_22555", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "tooling", "reproducibility", "security_gates"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f171040221edc5cc584ef9a066c3ccb4cb5395da"}
{"id": "train_22556", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "reproducibility", "security_gates", "documentation"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a48d85509cc98e6c0247c62339e59e73c05f87bd"}
{"id": "train_22557", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "auditability", "tooling"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b9ae39fc3210d9a99a9da7c3c712e4090fd88da6"}
{"id": "train_22558", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "tests_are_truth", "governance", "security_gates"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0aaf8fe9e46d7eed75631d2590fd677496ae6edb"}
{"id": "train_22559", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "ci_integration", "tooling"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8ff94c75f34d6585bfc3fee6cd18bd881e7b894b"}
{"id": "train_22560", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "governance", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1a071934e7bab6dc2ce5247c4ab355281c696354"}
{"id": "train_22561", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "governance", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "da9e4b43ff9f505782f14d3e7924cff5797616db"}
{"id": "train_22562", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "tests_are_truth", "reproducibility"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b88799c4bb0a6a412dd6db886cf96d130b6d7751"}
{"id": "train_22563", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "tests_are_truth", "documentation", "governance"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c0cfb1482740c00abcdccddde9b7c3d7ce13dc9c"}
{"id": "train_22564", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "evaluation_metrics", "documentation", "auditability"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5fde4917b7c9c737684f8fff02ef9ed17e3ec05c"}
{"id": "train_22565", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "ci_integration", "tooling", "documentation"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5b23210b8e1adb39d19d823abdae9c6676a2b0de"}
{"id": "train_22566", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "documentation", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5655d6467b3c0f5cbcdda0c8cae2616572965299"}
{"id": "train_22567", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "governance", "ci_integration", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "80e749d1221eb9171002559625f5da33c63f5e6d"}
{"id": "train_22568", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "documentation", "ci_integration", "security_gates"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dd542a6f8752a394242eacd1f5db614317f46ee5"}
{"id": "train_22569", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "documentation", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4ee4152892bdf7db648d2e40a249bd884fb10fbc"}
{"id": "train_22570", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "tooling", "ci_integration", "security_gates"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "aabc477eb2ca5c772d75b6593ac7b2b74f260b44"}
{"id": "train_22571", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "tests_are_truth", "ci_integration", "reproducibility"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ae5126d6b9cc3cef53c2a5a02e64f5605dcd13c1"}
{"id": "train_22572", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "ci_integration", "security_gates"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2c36e46e8398d28df87da9343d8964bfb9090158"}
{"id": "train_22573", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "documentation", "security_gates", "tests_are_truth"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "795698d3569748dea7a71f536cfd28f73d5639e9"}
{"id": "train_22574", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "security_gates", "documentation"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3908fd3e2d068d8b7416acbad8fbb808641585fa"}
{"id": "train_22575", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "auditability", "reproducibility", "governance"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "153e5b4e4b53b0d99082d072edee2c02ae3cbebc"}
{"id": "train_22576", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "governance", "ci_integration", "documentation"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "70358dbd5cb00f3f6e6cc90c86485d6d42c6db6c"}
{"id": "train_22577", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "governance", "documentation", "security_gates"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e4fbf551b056651f180bbc42ab2cc038cbb6a8c9"}
{"id": "train_22578", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "security_gates", "auditability", "governance"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dae6346f3423ece6be8c55454905db59369ed42f"}
{"id": "train_22579", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "cost_latency_tradeoffs", "governance", "tooling"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5e8c0772a9436d01ce445490e1675a82d1d14dfa"}
{"id": "train_22580", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "repo_scale_reasoning", "security_gates", "auditability"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3d363f35331700701dd423653455503e7ecb1a30"}
{"id": "train_22581", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "security_gates", "auditability", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "67fd529922195b4de120c728be807dc548ecc645"}
{"id": "train_22582", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "evaluation_metrics", "tooling", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "97f9b6cc4d9b002e187eec603d1c692a3a323fc0"}
{"id": "train_22583", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "evaluation_metrics", "documentation"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2f29a90ca7ca3241c29d23ac00fb800a8707a9ac"}
{"id": "train_22584", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "repo_scale_reasoning", "ci_integration", "tests_are_truth"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "42bd8db05e17f9e1af4aa2d8a538ea9b8e9f03f3"}
{"id": "train_22585", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "security_gates", "governance", "auditability"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fbaf1991db51564d4147aa26fdf65fcd28e613be"}
{"id": "train_22586", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "evaluation_metrics", "reproducibility", "auditability"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "163222f58d5143a89b2f6c26defe01447345f302"}
{"id": "train_22587", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "auditability", "evaluation_metrics", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "aa1db1df4d64f97f3ae6709757f75d686de161b4"}
{"id": "train_22588", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "governance", "repo_scale_reasoning", "tooling"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1979ff7244e1a4553418793392f21788322f93a7"}
{"id": "train_22589", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "ci_integration", "tooling", "documentation"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d8e1b0c9a9dbc03aecb9d4665853f4ba274aa9f9"}
{"id": "train_22590", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "documentation", "repo_scale_reasoning", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3f3e6b570d6ab12d12cbb9317b75df6421805f51"}
{"id": "train_22591", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "auditability", "tests_are_truth", "tooling"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f6b65663886f7204efe3c351ac8a3635bd0a9528"}
{"id": "train_22592", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "tooling", "documentation"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fd61884349c47211f41fa703220fc1cd94a1f02f"}
{"id": "train_22593", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "governance", "documentation"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ab5b44d3339290741258bfdd55697278c63ee2ce"}
{"id": "train_22594", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "evaluation_metrics", "tooling", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c7e62816c8fabed7748804a0effde87a11c10655"}
{"id": "train_22595", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "auditability", "tooling", "tests_are_truth"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d1dbff6846b77be6290edc577b6ab23c96beb105"}
{"id": "train_22596", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "auditability", "reproducibility", "tests_are_truth"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ec7119f2c49f0e2c0d6e22b855b7aa69c0d741b1"}
{"id": "train_22597", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "security_gates", "documentation", "tooling"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "514aabf2bab4880ad12c3bc7c9bfdcb7b0137682"}
{"id": "train_22598", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "repo_scale_reasoning", "evaluation_metrics", "auditability"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bcc0adc0e872f933044f7f4c96b9eddff947815a"}
{"id": "train_22599", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "auditability", "evaluation_metrics", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f2bd519b60c0afa8e1de002ba5fd748a622e1765"}
{"id": "train_22600", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "tooling", "cost_latency_tradeoffs", "governance"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7af723fd669218ebea717d8e745314344f2daab9"}
{"id": "train_22601", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "documentation", "governance", "evaluation_metrics"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "651527682d2c7a94bdeb1777098c7c73ab939849"}
{"id": "train_22602", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "security_gates", "governance", "auditability"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7c9a37c1a44c4bc826b7164022c96529ba6189f9"}
{"id": "train_22603", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "tooling", "ci_integration", "security_gates"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7bbb9f54329a0d9eda29395a850ee6d052b13edb"}
{"id": "train_22604", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "auditability", "governance", "tooling"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "825325a7b9cb644ec7d4da80d81b13238f6a1693"}
{"id": "train_22605", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "security_gates", "cost_latency_tradeoffs", "repo_scale_reasoning"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "30e4a821a1e28ecac122d37d0b16f390b2b6499f"}
{"id": "train_22606", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "documentation", "tests_are_truth", "reproducibility"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "74f22c009ce75feba2067981eb781fc988135ca6"}
{"id": "train_22607", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "evaluation_metrics", "ci_integration", "governance"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4ee1d9ed172a4a7d0f15d6f3744100232dfcbf9e"}
{"id": "train_22608", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "ci_integration", "auditability", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "67f2014e44227089797cd9e6612f3838728415fd"}
{"id": "train_22609", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "security_gates", "tooling", "evaluation_metrics"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "18bed592e4d0e7c49992e8d4ee9152f621c3a94d"}
{"id": "train_22610", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "governance", "auditability"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9255f279d0b419b025bc536b93500de4c15c17f7"}
{"id": "train_22611", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "governance", "documentation", "auditability"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e3378a8b1d2057b8ac23b23d815c1c1f197ac323"}
{"id": "train_22612", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "security_gates", "documentation", "reproducibility"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8542da109b398357abfe044e3862f73fe56ba7ac"}
{"id": "train_22613", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "auditability", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "87a304280921987ccabeab745e0af1a56a895506"}
{"id": "train_22614", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "reproducibility", "governance", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c4e25b1aec77004c5d52a83805808c0c9f69d4d2"}
{"id": "train_22615", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "auditability", "ci_integration"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "713c00474b71aa2623fd4512f4ba7cef854c5096"}
{"id": "train_22616", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "tests_are_truth", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a59d1485e43b705ed1cb3ff1660562efdecf2f45"}
{"id": "train_22617", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "documentation", "evaluation_metrics", "auditability"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "af4fa1a2d6c7a396907c41069674ef17fe9a738c"}
{"id": "train_22618", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "tooling", "documentation"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f780144cfc67154a6d95f59b7d6b2cca53b75678"}
{"id": "train_22619", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "tooling", "auditability"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f70c41d07eec8e1edb6056755a47540d43ee2173"}
{"id": "train_22620", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "tests_are_truth", "auditability", "reproducibility"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "401f5bf072879d7585db2b9855da360082772006"}
{"id": "train_22621", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "tests_are_truth", "auditability", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e183fe22439075b02dd35e4533fd571bc6f10a73"}
{"id": "train_22622", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["governance", "cost_latency_tradeoffs", "reproducibility", "security_gates"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7ad6fdc814968244ba22c59ebfb154ca325ceb46"}
{"id": "train_22623", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["auditability", "evaluation_metrics", "tests_are_truth", "documentation"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b4ab948815c9d6e152a8f15319849ea126035e9d"}
{"id": "train_22624", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "evaluation_metrics", "documentation", "reproducibility"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4d947711df35c9d0c4fb0d68c753f2040d4e9135"}
{"id": "train_22625", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "tests_are_truth", "reproducibility"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bee3f650345c67566576331d2e2070fc27cc07e9"}
{"id": "train_22626", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "ci_integration", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b4e51072e0e77dc1524daaf3b7a35806b66125fb"}
{"id": "train_22627", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "ci_integration", "auditability"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9e34cf14f84b4043cc9ddb29b3e073865bde434b"}
{"id": "train_22628", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "governance", "auditability", "documentation"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0fbfe7603e86439b1c738f79917948ac85100460"}
{"id": "train_22629", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "documentation", "ci_integration", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c61c1a861d459c2bc1aa727a2cb16dbf2bd2eb44"}
{"id": "train_22630", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "evaluation_metrics", "governance"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6bcdfc1d58fe6db1321835baaf7b22df9ed1ebcd"}
{"id": "train_22631", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "documentation", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1f537a09573328f91cea9118c14d70b423deb364"}
{"id": "train_22632", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["auditability", "repo_scale_reasoning", "governance", "tests_are_truth"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f12aad9b09169697646aa84a1382caa8381ae025"}
{"id": "train_22633", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "security_gates", "auditability", "governance"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9cc18ccbcaeafd189d21b53ba750120c3bd34222"}
{"id": "train_22634", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "tooling", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2c97e69915e68ea604aa41bdb856c4b6f1070e78"}
{"id": "train_22635", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "security_gates", "reproducibility", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "903d462fe18034272c3db757560eb30e56f59f87"}
{"id": "train_22636", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "ci_integration", "documentation", "governance"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ed096344c88ca9c5ab00dddf648338303c8c55e2"}
{"id": "train_22637", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "security_gates", "repo_scale_reasoning", "tooling"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8414c469959bb5adaf723dc682ef0d0949297b36"}
{"id": "train_22638", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "security_gates", "cost_latency_tradeoffs", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8205d7e274808f6ffa944ebfeb6131464f52c73a"}
{"id": "train_22639", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "documentation", "security_gates", "evaluation_metrics"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "826f1d8371bfe7366f46e7fd1a04b08250f25773"}
{"id": "train_22640", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "auditability", "tooling", "governance"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fde0599797bace6e4918f207f56a4ee43ff80c6f"}
{"id": "train_22641", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "tooling", "tests_are_truth", "documentation"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "33ce75fd7291f58f4ce6dca5bfa53af4be458c42"}
{"id": "train_22642", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "auditability", "ci_integration", "tooling"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bd0d0381ebfecfdbfab2155035d1647a5751c805"}
{"id": "train_22643", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "ci_integration", "tooling", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f26bd323a6429f451a1725762172c42baa45c1da"}
{"id": "train_22644", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "ci_integration", "governance"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4d6d5c19c7b0f0372de8a8d016b4b75a311dd21b"}
{"id": "train_22645", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["governance", "documentation", "auditability", "tooling"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ed84b4a9879b194e3b6c15ed27c31c3eaa48099b"}
{"id": "train_22646", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "reproducibility", "auditability"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e26c96ce4caf48375bd5f30c4866c7d8e0e0133a"}
{"id": "train_22647", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "governance", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1ab2bab09b698268c953007277d4a3111ffd04d2"}
{"id": "train_22648", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "ci_integration", "governance", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dd298fd15ba61c7dc8b967a256c9f4c31305c638"}
{"id": "train_22649", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "reproducibility", "governance", "documentation"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b6c8036cc389404d23adfa7dbd8edda1b81f6884"}
{"id": "train_22650", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "documentation", "security_gates", "tooling"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "90434df9c9fc4f4a4c2ba1952f3a9288f455ef41"}
{"id": "train_22651", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "evaluation_metrics", "security_gates", "governance"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1d72872900be1160437eef7dd4bd6a9ccef48e12"}
{"id": "train_22652", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "reproducibility", "evaluation_metrics", "documentation"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9dbb9c0a0a3790af549124ebebb050295e992ca6"}
{"id": "train_22653", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "documentation", "ci_integration", "security_gates"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fbc935c9a8632f3d657d840b3f626eca9f16c5d6"}
{"id": "train_22654", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "security_gates", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c6a4420c3abaf22b63d2e5661d64dbf73c120728"}
{"id": "train_22655", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "tests_are_truth", "evaluation_metrics", "repo_scale_reasoning"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1d4ec445ff4eb880877408461c9bedc459b50b75"}
{"id": "train_22656", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "tests_are_truth", "reproducibility"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e6964c961a25298773032a46f27da07a434d9d55"}
{"id": "train_22657", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "auditability", "security_gates"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "512b8bd43a126e6758ac4c47fa6e67bc70f25eec"}
{"id": "train_22658", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "documentation", "tooling", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f593188eaf1f3ad4120fe7e8b80387e60fb935dc"}
{"id": "train_22659", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "tests_are_truth", "security_gates", "reproducibility"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ac2d0f906a277b80239bdacbe4248c6b3b80056e"}
{"id": "train_22660", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "security_gates", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7823147a4a63257df8da6a46aa028bc1132272b7"}
{"id": "train_22661", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "auditability", "governance", "tests_are_truth"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "666224b83856cc29651ee76820c7a79dc30017d4"}
{"id": "train_22662", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "security_gates", "auditability", "reproducibility"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ae7a251b74119b39e05930da5634ed1dd3eb287b"}
{"id": "train_22663", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "evaluation_metrics", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "10ed97d645168d440dfbd9e2a95a6808df94dc87"}
{"id": "train_22664", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "documentation", "tooling", "auditability"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4cbb10c0793df64c1439eb44574bde2021e11a0d"}
{"id": "train_22665", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "governance", "tests_are_truth"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2518002a55ca5242c8952cdc167f2a9492a728d8"}
{"id": "train_22666", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "documentation", "evaluation_metrics", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "08ef84efc15aae3b33da8c034f6319c05bf1dc00"}
{"id": "train_22667", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "documentation", "security_gates"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7e540200dccc6dccf6e8bba771bc393cac628702"}
{"id": "train_22668", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "evaluation_metrics", "tests_are_truth", "ci_integration"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e45528700ba52ecbdb4a67b1f8d141e13dad2876"}
{"id": "train_22669", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "governance", "documentation"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7dae18790b7006890d81afd26cac4c0d1dd85e5c"}
{"id": "train_22670", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "reproducibility", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4628b463ca6a385509cc99acd87649f9401b3063"}
{"id": "train_22671", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "reproducibility", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f59d980016e819ad9c844f914cd8b7d7794a9b79"}
{"id": "train_22672", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "reproducibility", "documentation", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6e46a25b0445b655c98f2f3590e85a5a6b805102"}
{"id": "train_22673", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "ci_integration", "governance", "security_gates"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c608ad398cfeb5e2d41f2b1d2ffcd36489a786c9"}
{"id": "train_22674", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "documentation", "auditability", "governance"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "27152c392ae9c5cb24d5a4c1dc06be3a08331d72"}
{"id": "train_22675", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "documentation", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "34a5fd2b09c42bc5cc0105ecdbe02ac8582d5647"}
{"id": "train_22676", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "documentation", "tooling"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "862510931fd7dc629654679fdcc7ee26bbde4a5f"}
{"id": "train_22677", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "ci_integration", "repo_scale_reasoning", "evaluation_metrics"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a44c376efd1863fc74d0bd8c1a2697460ce25918"}
{"id": "train_22678", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "tests_are_truth", "repo_scale_reasoning", "security_gates"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1c9e1515bd2c1e2596491460aca7083b66d20b18"}
{"id": "train_22679", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "auditability", "tests_are_truth"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3c38960e818a3047423dfb55b8f9caddb8206fe6"}
{"id": "train_22680", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "evaluation_metrics", "tooling"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ebd29f915209c4ff0e4436b8c549a49b4da32b97"}
{"id": "train_22681", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "evaluation_metrics", "tooling", "auditability"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "af9258a6ba33f67cee363b92f3e39962e977206c"}
{"id": "train_22682", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "governance", "tooling", "reproducibility"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "80fce0ad966adcb7ff6df048d6a053024f8b3785"}
{"id": "train_22683", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "tests_are_truth", "governance", "ci_integration"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0d1b0f464d8d520dcaabefc6a4231e8f732a1e3d"}
{"id": "train_22684", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "security_gates", "governance", "documentation"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5d05778590b94f6bc360324b68f930383feeb70c"}
{"id": "train_22685", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "governance", "documentation"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d40f60074f5c1f3ac01fee6fe161189796e22ca0"}
{"id": "train_22686", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "documentation", "tests_are_truth", "tooling"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1be04c48c8332cda3d12713721cdbb2ca3bb05bc"}
{"id": "train_22687", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "reproducibility", "evaluation_metrics", "security_gates"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ef8c3e27501c58a1916e70ff498fdf9f54c0c912"}
{"id": "train_22688", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "governance", "ci_integration", "reproducibility"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9417e8e80f8fc6b070a268e32559415150de12b1"}
{"id": "train_22689", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "documentation", "ci_integration", "tests_are_truth"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7f8cf9c2c9ed1c4a0db82f1d26271d45617ef685"}
{"id": "train_22690", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "repo_scale_reasoning", "ci_integration", "security_gates"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "55e7003b115a0acf58dec68f75d1d67a929894e9"}
{"id": "train_22691", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "tests_are_truth", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bd2c361c94acc4ae5eecb2464b855231b56da070"}
{"id": "train_22692", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "security_gates", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e66d24de0d2ee733c85aa639885eb37bfabe1f79"}
{"id": "train_22693", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "evaluation_metrics", "reproducibility", "documentation"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "da40e6d646845b3b9bd36e1b3dc23e567dbbfbe9"}
{"id": "train_22694", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "auditability", "cost_latency_tradeoffs", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a78aa6f36e6ef3fbe43bb6affc4297823754e70a"}
{"id": "train_22695", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "tooling", "ci_integration", "reproducibility"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2b159156b86bf87c4b2aa608faff125ab0410cce"}
{"id": "train_22696", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "repo_scale_reasoning", "reproducibility", "tooling"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "92a844b97a26997d06aa52976c61ba7ed4928835"}
{"id": "train_22697", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "governance", "evaluation_metrics", "documentation"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d57062eab08fbbf2aa0716137c0936f1bd9cdba4"}
{"id": "train_22698", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "tooling", "evaluation_metrics", "tests_are_truth"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a5cd725038e9be7984fdfb22ddf5f14a6e82bff9"}
{"id": "train_22699", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "tests_are_truth", "evaluation_metrics", "security_gates"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3218ee57c5faf33dda30d08b0617088ebf55d146"}
{"id": "train_22700", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "tests_are_truth", "auditability", "tooling"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "102c9f6e44d378faf9e5e9f515813f73defefd45"}
{"id": "train_22701", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["auditability", "tests_are_truth", "reproducibility", "evaluation_metrics"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "08e7f11b4e56d8b6df3a5bfa6a763bdfd6e8a2ae"}
{"id": "train_22702", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "tests_are_truth", "auditability", "security_gates"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "55c8a16f109c277b4cdc19ed2571a079e1a7dfa9"}
{"id": "train_22703", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "governance", "repo_scale_reasoning", "tooling"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0845d7f12152810a0096d7b025299e30ebf388e3"}
{"id": "train_22704", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "ci_integration", "documentation"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b853ca0a6eecac6e61cfc7effe1875ca682d8061"}
{"id": "train_22705", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "security_gates", "auditability", "tooling"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7e9515de35805a87bd40c4791d916bd62fca00fa"}
{"id": "train_22706", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "tooling", "governance", "reproducibility"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c3a66699c6a7b0a7925f18eba43860d86f9dc90b"}
{"id": "train_22707", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "security_gates", "repo_scale_reasoning", "evaluation_metrics"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b539e6f38883ab91d86f725984ab5e83aaf4dd7d"}
{"id": "train_22708", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "reproducibility", "repo_scale_reasoning", "governance"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0a671a4b3f5b4df36b9ca9531922fdfc1304ce27"}
{"id": "train_22709", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "ci_integration", "tooling", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "be25c349dc8b3e5115af40334df25e776c75c9ee"}
{"id": "train_22710", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "governance", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "018f1fa145a75af5831f7622c6146a06f7fd8575"}
{"id": "train_22711", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "auditability", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8838b9665a90f082a348223bcb12c688bb946153"}
{"id": "train_22712", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "tooling", "governance", "ci_integration"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8e7f3e633e47dfeea1941aee4448f632d090eadd"}
{"id": "train_22713", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "evaluation_metrics", "reproducibility", "governance"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "37ba3767890ffb9c6c5409240ee56fb212e38f2f"}
{"id": "train_22714", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "cost_latency_tradeoffs", "auditability", "ci_integration"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f7668d5e0604efa14053d01c835b8a1ce9b117a3"}
{"id": "train_22715", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "security_gates", "reproducibility", "auditability"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9b0bcf06825c4ebee4d342389322a9e5ea756803"}
{"id": "train_22716", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "tooling", "evaluation_metrics", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bad73bb0f5691fef02438bdb90fa4ffc2ad02413"}
{"id": "train_22717", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "auditability", "tooling", "documentation"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b68f7f72a45cfa93c55eaba80c563ab58a29e43b"}
{"id": "train_22718", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "ci_integration", "governance", "security_gates"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2316773b59921a6b4f068bcafea4895e2b49b59b"}
{"id": "train_22719", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "tooling", "auditability", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8984656bb921cc2ab16352328f2869e7c2eeb71f"}
{"id": "train_22720", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "tests_are_truth", "governance", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7a09d3631c198f634697916c0eb43bbcf5038c5f"}
{"id": "train_22721", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "tooling", "tests_are_truth", "reproducibility"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ae6e829973e052a5776f6a0c11c6ed7aef5d70df"}
{"id": "train_22722", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "security_gates", "tooling", "auditability"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a9e65f5881bf33b51c85e5c37f489b7d544a4327"}
{"id": "train_22723", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "tooling", "auditability", "governance"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "77f433e416f6d40c16f24469f80ae7c8f7b0794b"}
{"id": "train_22724", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "auditability", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2002cb30ece360646429654a6e1256e15aad1c35"}
{"id": "train_22725", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "governance", "reproducibility", "evaluation_metrics"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ee704df96fdfd5c4365fef95660d400229db4c5a"}
{"id": "train_22726", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "governance", "security_gates"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "abae2346d2f87f83f893e7b9bf8c56676607bb2a"}
{"id": "train_22727", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "evaluation_metrics", "tests_are_truth", "tooling"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f93b3ae0c9a3fb276209830924d3f441d769b1c0"}
{"id": "train_22728", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "governance", "repo_scale_reasoning", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "73101921169c5ce7b7673407e4e62e000d5db79c"}
{"id": "train_22729", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "cost_latency_tradeoffs", "tooling", "auditability"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f621e30bd75e0958c61a42d5a4b8627f8c1219be"}
{"id": "train_22730", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "evaluation_metrics", "governance", "security_gates"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3d2dc9c7f49e512cea4ac438300bd4b3fef34a0f"}
{"id": "train_22731", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "tests_are_truth", "evaluation_metrics", "tooling"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ccdc9fd397e9189af5524c891ad92a12b04e59dc"}
{"id": "train_22732", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "evaluation_metrics", "reproducibility", "governance"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f9c43397811f277df0c933b5b50c4742de3191fb"}
{"id": "train_22733", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "security_gates", "reproducibility"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fdbb21ad52c18e4d8a2354f51377c79572f02b58"}
{"id": "train_22734", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "repo_scale_reasoning", "governance", "tooling"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f3deefd1ff7fbd53b691f5206fb1a3e0bc00071e"}
{"id": "train_22735", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "ci_integration", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e67274a16372075bc733bf63731830dfc2f6a765"}
{"id": "train_22736", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "evaluation_metrics", "repo_scale_reasoning", "governance"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "01c681b1a5de98d750e2e4c82232df6113595ef2"}
{"id": "train_22737", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "tooling", "documentation"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b9af73cb16f3f225f514c160b71e9d1a5b8df4e4"}
{"id": "train_22738", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "ci_integration", "tooling", "evaluation_metrics"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b76a148e23237b63d2e9d6088429c9a20b7068c8"}
{"id": "train_22739", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "tooling", "documentation"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c16b1537e3fff653656886edf5635f2f96f5c272"}
{"id": "train_22740", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "reproducibility", "documentation", "tooling"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ae95c77ece7c20695ef27c2b6b84695ae923313a"}
{"id": "train_22741", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "governance", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8dea5d220a0ffa467db3a811aeae6cdee0b9631b"}
{"id": "train_22742", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "ci_integration", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bd56a260359f4b8f64bc68af3668880d171ee71d"}
{"id": "train_22743", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "tooling", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c5ce68aace23aec509eaf1079d235ce993ebb37f"}
{"id": "train_22744", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "governance", "documentation"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c629afe13c55743c427219b82dd3fe0a040e460a"}
{"id": "train_22745", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "evaluation_metrics", "documentation", "ci_integration"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c71d8bac33cd0ac52721b6ecf02d02967f8f4fd5"}
{"id": "train_22746", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "reproducibility", "governance", "tooling"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a5bb88c403efbcf4b537a1f1d41dd0404c06e044"}
{"id": "train_22747", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "documentation", "evaluation_metrics", "ci_integration"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b25f6a851ba7e950eb6d743baf99a4543a2813e7"}
{"id": "train_22748", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "auditability", "security_gates", "reproducibility"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f2dc2ebef4f545f2420338455f4d257999ddae66"}
{"id": "train_22749", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "governance", "security_gates", "tests_are_truth"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0d8261f06f855e78b2485de470a05d2026bfa427"}
{"id": "train_22750", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "documentation", "auditability", "tests_are_truth"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "264aa07c9b9b6e5ee6471c3c5bf019a98ddcb0ee"}
{"id": "train_22751", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "auditability", "repo_scale_reasoning", "documentation"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "66575a0ad0d888fa8fe3ad005a03fc915ca5775a"}
{"id": "train_22752", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "cost_latency_tradeoffs", "security_gates", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1c3ff544145b28b4c249a768ce97dc3f3bab0082"}
{"id": "train_22753", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "ci_integration", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "81fce9a22932d3a1a3f5c4cd4b60d4e2d641aaaa"}
{"id": "train_22754", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "tooling", "security_gates", "evaluation_metrics"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "43d8e8c11339bc0a38b05cf4ca7cc05ea105fd60"}
{"id": "train_22755", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "cost_latency_tradeoffs", "tests_are_truth", "ci_integration"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "006c8046b4a14736666f198ce3830499e3c9a38b"}
{"id": "train_22756", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "governance", "reproducibility"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b49df7a7c521b798d3af2cc6a3a195e8493ff802"}
{"id": "train_22757", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "evaluation_metrics", "reproducibility", "ci_integration"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7b07d810b8fa60cd15f0e869d95c387a4d504e9d"}
{"id": "train_22758", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "security_gates", "repo_scale_reasoning", "documentation"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "10e6bc6d048e41a011fa69c2b8d167c614e62b46"}
{"id": "train_22759", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["governance", "ci_integration", "tests_are_truth", "reproducibility"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "87e39b05d6506fd6dbb920befe2bb668caeee720"}
{"id": "train_22760", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["auditability", "repo_scale_reasoning", "reproducibility", "security_gates"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f04b3160916a98ee9f09f7f08bda8645cfa022db"}
{"id": "train_22761", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "tooling", "documentation", "security_gates"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "17b759c8f43062cb20b086efee53f63da9867f83"}
{"id": "train_22762", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["governance", "cost_latency_tradeoffs", "tests_are_truth", "ci_integration"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "059108832d5b5f5ccf3c09d247e51750be3512c6"}
{"id": "train_22763", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "tooling", "tests_are_truth"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "941bb3e6de7f4cb469d2b87858adb4a8a74e942b"}
{"id": "train_22764", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["governance", "auditability", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a490765279bb9f6a4f9d63e5d28b021052cad560"}
{"id": "train_22765", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "governance", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b75e83242d6a8f024c166f7cbe9b9e1dbd693335"}
{"id": "train_22766", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "governance", "tooling", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "deaaa043fcd1af2724abbeeaf1865918f59b7f4d"}
{"id": "train_22767", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "ci_integration", "governance"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e0eb4d2a36e926155e848c3976a950190843b97a"}
{"id": "train_22768", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "evaluation_metrics", "documentation"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5dde7913c23b14596decebb862ce8e3fc0e4c5eb"}
{"id": "train_22769", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "ci_integration", "repo_scale_reasoning", "security_gates"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f81cef50d0962ec1828ceed32be6585aafbf9433"}
{"id": "train_22770", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "governance", "repo_scale_reasoning", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9bc93c7ffb9eb2eec10a08a35e688e25931a9b14"}
{"id": "train_22771", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "governance", "tests_are_truth", "tooling"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "779e9d7b451cbc3aefb20643e09fff850f1561cd"}
{"id": "train_22772", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "governance", "ci_integration", "documentation"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1d3dd395b00dbc696282980bb3b1ccee90daf49e"}
{"id": "train_22773", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "documentation", "tooling", "security_gates"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3d09cec57ae79c362a9986882402146d1858b5cc"}
{"id": "train_22774", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "tooling", "security_gates", "tests_are_truth"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6ea246326f559030a5ac78deaaae3ba0174ba559"}
{"id": "train_22775", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "security_gates", "governance", "ci_integration"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4708e0b2bc32a49c3c79170cb4078cc629a9b786"}
{"id": "train_22776", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "auditability", "governance", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fedd81a6a619330a685d9dd52e1a9be9b74bffa0"}
{"id": "train_22777", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "auditability", "reproducibility", "evaluation_metrics"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4e57c4265c8c7844e06cd0577c05cde1e5518983"}
{"id": "train_22778", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "documentation", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c4293ecfccd76480f0a0d53170bd73da9a603b31"}
{"id": "train_22779", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "tooling", "security_gates", "reproducibility"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "522c8569dc035f74742e008a1c5ee90c4e45e676"}
{"id": "train_22780", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "security_gates", "tooling", "tests_are_truth"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "415e4c9ae19e12a7a94ec631fd2a63de2fa4aaf7"}
{"id": "train_22781", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "governance", "ci_integration", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a1a6ab57b6965d1f33aa13d44bcb283927103965"}
{"id": "train_22782", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "tooling", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cb17d87b8e9aaaca01d32b8b2577af9f76432982"}
{"id": "train_22783", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "ci_integration", "auditability", "reproducibility"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "61e916c6b463deaa45e535c50f67dad6c16a5479"}
{"id": "train_22784", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "security_gates", "auditability", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "81600ff28cac15c01bff222dea1ac235f6ecaee1"}
{"id": "train_22785", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "ci_integration", "security_gates", "repo_scale_reasoning"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4fae9208a6b3c788a6db68485c27ad253eef15e1"}
{"id": "train_22786", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "evaluation_metrics", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2e312535c26f2780917f94d6420b435e2e06eba7"}
{"id": "train_22787", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "tests_are_truth", "reproducibility"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d6df8933713418ba5bcf5577cbd4b42a57975134"}
{"id": "train_22788", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "cost_latency_tradeoffs", "repo_scale_reasoning", "governance"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f6a94db77ffe89a51ccf20630934729332241aa4"}
{"id": "train_22789", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "reproducibility", "cost_latency_tradeoffs", "governance"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "569ac8b73773cf411997774267d47d4f073d0f4e"}
{"id": "train_22790", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "security_gates", "reproducibility"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "82e5b08df66b86992d0749d22a660a41cf5048f4"}
{"id": "train_22791", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "evaluation_metrics", "auditability", "documentation"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "672e9f4a5f927846dc1669585d9676175bc8aebf"}
{"id": "train_22792", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "ci_integration", "documentation", "auditability"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ff33eef93bda1de75c1b1bfc4cfe6920c1f58ee4"}
{"id": "train_22793", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "auditability", "tooling"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9fc43ec017781ec6b4c29717bd591531b73b75b6"}
{"id": "train_22794", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "security_gates", "tests_are_truth", "tooling"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bb09ab70ad1b2e40a622c4626630b7cff58f5714"}
{"id": "train_22795", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "auditability", "reproducibility", "tests_are_truth"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ae8c3874f1ee653a3d5475933dc1979bef5bd254"}
{"id": "train_22796", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "documentation", "governance"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f12d25d897ba2650fcdb13e7e9c5c13c05aa736c"}
{"id": "train_22797", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "auditability", "documentation", "governance"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "15d222a9a167b1b4d13d5ecbe35bf269eb906582"}
{"id": "train_22798", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "governance", "tests_are_truth"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "60deb7e1f227ae109d5fae9b0a8427eccc1c5917"}
{"id": "train_22799", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "cost_latency_tradeoffs", "governance", "ci_integration"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "605a9d35fd76ce2c9748fd0fecca6fbf59d10e96"}
{"id": "train_22800", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "ci_integration", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9af7f6b1dafcd90ee034d9272a915f33ffe2e885"}
{"id": "train_22801", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "evaluation_metrics", "reproducibility", "documentation"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cc72c192bae800a8215212cc82b57ea7c09ea2a2"}
{"id": "train_22802", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "ci_integration", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0fbf83c4e6e0bc2b06cca3e8cae52ad4d3a4ac2a"}
{"id": "train_22803", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "cost_latency_tradeoffs", "evaluation_metrics", "security_gates"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "555b67e35006f97b24b5aabc982a1e65262fc047"}
{"id": "train_22804", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["governance", "security_gates", "auditability", "tooling"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f26ab9416b7d5cdc28cab99771262e8f293469ec"}
{"id": "train_22805", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "evaluation_metrics", "documentation", "governance"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8a72187bef4a32e9da7b609593101ce7719f3275"}
{"id": "train_22806", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "ci_integration", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8577b127544c50f9d885ea54f68dc407930425e7"}
{"id": "train_22807", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3995765778210767e00f68632e4309d4b98141f0"}
{"id": "train_22808", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "tooling", "repo_scale_reasoning", "auditability"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a82c04c457636f32d1be3e4eb7c3d6db287c9476"}
{"id": "train_22809", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "evaluation_metrics", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8f6b9093a92d9970e967ce6eee8cbebe7e607874"}
{"id": "train_22810", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "cost_latency_tradeoffs", "documentation", "tooling"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "70b11b84be0213249f39ff2d20b2e07372c4a91f"}
{"id": "train_22811", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["governance", "cost_latency_tradeoffs", "ci_integration", "security_gates"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a9e65f5881bf33b51c85e5c37f489b7d544a4327"}
{"id": "train_22812", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "ci_integration", "reproducibility", "documentation"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d39fd726e37eaa3fb985f89d4ff619c5b3d9f1e3"}
{"id": "train_22813", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "repo_scale_reasoning", "reproducibility", "documentation"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "117d5aa3bb53de119878b3ebc8efdf8727a4b2d4"}
{"id": "train_22814", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "reproducibility", "tooling", "documentation"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0bae9704113f2be2322c1d11f705c59e539ccce3"}
{"id": "train_22815", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "reproducibility", "tooling", "tests_are_truth"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "981cf04e5e76c538f039a3a2ced8503d3db90d09"}
{"id": "train_22816", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "tests_are_truth", "security_gates", "tooling"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d160c81e40bb990a876632aa751860372af063b7"}
{"id": "train_22817", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "governance", "repo_scale_reasoning", "security_gates"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c871fe9827c39203af43beb3939732cd7b900c64"}
{"id": "train_22818", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "security_gates", "evaluation_metrics", "reproducibility"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a3a164268219e8a7381bba7ba7037b9b7cb35243"}
{"id": "train_22819", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "cost_latency_tradeoffs", "tooling", "ci_integration"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6c275e4cc3411d402cdc9ce7a689ea02ede5208c"}
{"id": "train_22820", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "security_gates", "documentation", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bb1afe56c68709727227007a8d70f384b4705c22"}
{"id": "train_22821", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "repo_scale_reasoning", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a00f06c039afc557a7be3a8d8be8490223bfd67b"}
{"id": "train_22822", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "security_gates", "cost_latency_tradeoffs", "governance"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a53150fc9fea421c3bef73e76f9adf88d5944ca1"}
{"id": "train_22823", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "cost_latency_tradeoffs", "repo_scale_reasoning", "evaluation_metrics"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e622f47194e0036d9be983607368c2707edc872d"}
{"id": "train_22824", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "evaluation_metrics", "security_gates", "governance"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8e17c46c2129e33cedbe3da27ee634144f7a3f58"}
{"id": "train_22825", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "89acd939dc5cf1c5d5c04c85484686cdf5b3fd57"}
{"id": "train_22826", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "auditability", "reproducibility", "governance"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5d3c8b50b80176bb3c1b17922401e3c08f48f40b"}
{"id": "train_22827", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "reproducibility", "evaluation_metrics", "tooling"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e08935c44316cf992f7d08743aedf5d09f7609b2"}
{"id": "train_22828", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "documentation", "reproducibility", "auditability"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cd394ce393bf5f59f0f2906b8d66ca2811b80937"}
{"id": "train_22829", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "security_gates", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d4912312a2dce020b4c1d6871f64edb53719f63a"}
{"id": "train_22830", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "evaluation_metrics", "reproducibility", "documentation"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f3ceb2932d1c9323f2bca6d496205e1449a62464"}
{"id": "train_22831", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "tooling", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f7e75ff58c5324d75b3962a40bcc1060218aca41"}
{"id": "train_22832", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "auditability", "governance", "tooling"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "857a18985a94bb9ff8af8adc34a94465a62c649c"}
{"id": "train_22833", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "governance", "reproducibility", "tests_are_truth"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "836f44fa1532d88a0aa593b7af7524f880776d68"}
{"id": "train_22834", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "ci_integration", "security_gates", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c59e0b93d367bd52f8bcd87dd56d5b29c08a0bb0"}
{"id": "train_22835", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "documentation", "governance", "ci_integration"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "da253274444462a3a56bbf0acabf9ce6a576f445"}
{"id": "train_22836", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "security_gates", "evaluation_metrics", "reproducibility"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "53bf97f86b1ef481548b2510a1414538f889b3b4"}
{"id": "train_22837", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "governance", "tooling", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5d6ddcb3bf19a13f8cd102fee0d2c7e6e60d3847"}
{"id": "train_22838", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "repo_scale_reasoning", "auditability", "documentation"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e3e01ed2c3de34d806b4047a26ffd52448fdacaa"}
{"id": "train_22839", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "tests_are_truth", "reproducibility", "governance"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "92076bfa4879a3599b4aff6ffde635d96d4034ae"}
{"id": "train_22840", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "tooling", "documentation", "reproducibility"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "59cc8568fc4b2eb715fa7f4e81462f52f3247c9a"}
{"id": "train_22841", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "tooling", "governance", "documentation"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2ae8305c3b7919e956f37dbd3ed81dc7f75dccc6"}
{"id": "train_22842", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "documentation", "governance", "evaluation_metrics"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c96b131c334632608d26b76b5cbac113ba2c1a69"}
{"id": "train_22843", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "governance", "evaluation_metrics", "tooling"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8c674d767f439b31a8f6738208ec6561cbd63161"}
{"id": "train_22844", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "documentation", "evaluation_metrics"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e6830f6d480626b865fe063c298d38aa9581faaa"}
{"id": "train_22845", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "ci_integration", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b0d17acae5b00b8d6eea8c5ce14e48b29d98dfe3"}
{"id": "train_22846", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "tooling", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "27104f9f9ba002dc3348abdaeadd636b02e385c8"}
{"id": "train_22847", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "tooling", "auditability"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "66ba4da29b7fe3eeacb6aa113d36b5c48059ffe6"}
{"id": "train_22848", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "cost_latency_tradeoffs", "governance", "evaluation_metrics"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7b510919f7f3cd3b8759d5d49ddd39e15c94dcb3"}
{"id": "train_22849", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "evaluation_metrics", "reproducibility"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a8a23a4050ee782d8a3539acfcf807d94fb7820e"}
{"id": "train_22850", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "reproducibility", "governance", "ci_integration"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cc73c4b27ef6b31b2ec082bfc97635c3a8b349d4"}
{"id": "train_22851", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "security_gates", "tooling", "tests_are_truth"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a969d64959ca233183dcd9baddaaa832b62f0739"}
{"id": "train_22852", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "governance", "repo_scale_reasoning", "documentation"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0f2560a2f23b0e7cd7c6c423ad44048366d070cb"}
{"id": "train_22853", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "ci_integration", "governance", "tests_are_truth"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cbcb5cbb316d1e02cd9adbae30a1eed5e358ace6"}
{"id": "train_22854", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "reproducibility", "tests_are_truth"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "080982ea5b1ebc0ff7c54c33d881461131c7d6b4"}
{"id": "train_22855", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "security_gates", "governance", "reproducibility"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8e8724f40ec4108f2715c346b172cbf66b996564"}
{"id": "train_22856", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "tooling", "ci_integration", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "042ebd0002c0501397376d40132beda486cbe319"}
{"id": "train_22857", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "security_gates", "governance", "evaluation_metrics"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6bf819ba8871a82b9c0ea5738d0323a1cdfebe6c"}
{"id": "train_22858", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "repo_scale_reasoning", "cost_latency_tradeoffs", "governance"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "314b0b11efc11918408b5b32a96dceeda4aeb32a"}
{"id": "train_22859", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "evaluation_metrics", "ci_integration", "documentation"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "504d7a943f184a5635cfcc78f1e2babc010b6541"}
{"id": "train_22860", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "ci_integration", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "804be2199a690299ff6ae970600f75f6d7498d64"}
{"id": "train_22861", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "repo_scale_reasoning", "tooling", "security_gates"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "51fe12ec33ba690b357503d4064392ce97e11145"}
{"id": "train_22862", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "security_gates", "tooling", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f5f48f8b82febcf7e3e1f5ae3d43c863a5414fca"}
{"id": "train_22863", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "auditability", "security_gates", "documentation"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dd2bf051c9b9d99c9413d17ef1e393ceb4be08b4"}
{"id": "train_22864", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "cost_latency_tradeoffs", "evaluation_metrics", "documentation"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4b1fb289bf80467a51ea3fb20ebd8819e6412093"}
{"id": "train_22865", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "reproducibility", "ci_integration", "tooling"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2b8c535d2afecf6071dd5e8ee9aa6ec63deeb29f"}
{"id": "train_22866", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "auditability", "security_gates"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a32a26cd5d71fa7421068aa662fbfdcb6dd5ef88"}
{"id": "train_22867", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "documentation", "ci_integration", "auditability"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "76a46615dcc12079f79e617b651d1a619c0611a4"}
{"id": "train_22868", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "ci_integration", "governance", "documentation"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e0a4b725e4fc90b38061e77983957f58f4f9eca2"}
{"id": "train_22869", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "tooling", "ci_integration", "reproducibility"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "460df273a8cd7744bc56dd28c8237deb7106d7ff"}
{"id": "train_22870", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "documentation", "reproducibility", "security_gates"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "354b33c143e4ac2a00b7cbecb7d8b5d9bf959f7f"}
{"id": "train_22871", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "tests_are_truth", "auditability", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f21d936857101d14163880b600a0415a4fa77797"}
{"id": "train_22872", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "tooling", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "39ffb8c98aa2374368f300b1c73d6ee9d6c8bacb"}
{"id": "train_22873", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "tooling", "governance", "tests_are_truth"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ee3be44609c1be7e06fa1ba006ebff359d1bf56e"}
{"id": "train_22874", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "tooling", "ci_integration", "evaluation_metrics"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0f53be0b5288d2a747575a5b78eafa08dcd53a47"}
{"id": "train_22875", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "evaluation_metrics", "ci_integration", "auditability"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "918d0744b72cdad74711c1391fef3aa2704f18e9"}
{"id": "train_22876", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "auditability", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cf6003dff70abf96796778c9a6724cf3a6e4baa7"}
{"id": "train_22877", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "tooling", "evaluation_metrics", "documentation"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "54de4d6d318669347357ff35cc9bb08e6b74df20"}
{"id": "train_22878", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "security_gates", "documentation", "tooling"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "27beaaeeab855d2314c36ff885a62c77f792b645"}
{"id": "train_22879", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "repo_scale_reasoning", "tests_are_truth", "security_gates"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f566b71874b6ba09033e7061cb5de2a1ce2d00ed"}
{"id": "train_22880", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "cost_latency_tradeoffs", "security_gates", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6ce9f87a599d6e5b34f1ad92df8fa76c35d12338"}
{"id": "train_22881", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "repo_scale_reasoning", "cost_latency_tradeoffs", "security_gates"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c813e9869ee4d546939b6bc59e7434ccea18ac89"}
{"id": "train_22882", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "cost_latency_tradeoffs", "reproducibility", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "974eb9bb0e474efa52d7ba6bd689913fe242015b"}
{"id": "train_22883", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "tooling", "auditability", "ci_integration"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "127854640f009b147b84edbf373a38a71b5b2476"}
{"id": "train_22884", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "documentation", "security_gates", "tests_are_truth"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ef6fcbeacdf676c062849eb0eca205e07ee82542"}
{"id": "train_22885", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "security_gates", "tests_are_truth", "tooling"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e538feef7b8e008605d2d66da57a8c9334bc933d"}
{"id": "train_22886", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["governance", "security_gates", "documentation", "auditability"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "19af14e3490976295aea01c4f7b1aec3a29fa722"}
{"id": "train_22887", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "auditability", "repo_scale_reasoning"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "026210b43ad7cd34b632692c3ccdc7af69215201"}
{"id": "train_22888", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "evaluation_metrics", "documentation"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "305ab2356b5ecfaf2a798c8107c0dd075366bf08"}
{"id": "train_22889", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "documentation", "evaluation_metrics", "reproducibility"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "54e134a00b8de70d94962bbeeef60003e717d248"}
{"id": "train_22890", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "security_gates", "auditability", "tooling"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "10c07a9440493a2db532665880287c523cc9b313"}
{"id": "train_22891", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "repo_scale_reasoning", "reproducibility", "documentation"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fdfbf2bb23f99fa0c009ff06441a9b6077b1dff8"}
{"id": "train_22892", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "auditability", "evaluation_metrics"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7b917e4c0e2ded63401d46eb7432c2c5ea58b49e"}
{"id": "train_22893", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "tests_are_truth", "governance", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4ed0f2c3c68ef84f2a6df5044dc6f2683be482f4"}
{"id": "train_22894", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "cost_latency_tradeoffs", "evaluation_metrics", "documentation"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e48abd6c5fb4bb0c1d7e98efa041c22ac0ec01a8"}
{"id": "train_22895", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "documentation", "ci_integration", "reproducibility"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "44e21cdad21ea2313df2078dacf18e1eef9c7214"}
{"id": "train_22896", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "tooling", "auditability", "governance"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1f25f15f3e4b0232509f8c6957ad96fcf32e4a96"}
{"id": "train_22897", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "cost_latency_tradeoffs", "documentation", "tests_are_truth"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cf89c0ca4101522974a13ce60c4a8132bb0ff251"}
{"id": "train_22898", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "tests_are_truth", "documentation"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a4dbf7b19c14c99197d85c7fd7870968237afddc"}
{"id": "train_22899", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["auditability", "reproducibility", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b8764705f751db6c529c210f10fca2da487e4c93"}
{"id": "train_22900", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "cost_latency_tradeoffs", "tooling", "tests_are_truth"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "26bb7349ab6e0f3b7872cad5869d1565c14454d7"}
{"id": "train_22901", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "tooling", "governance", "documentation"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6bf35917f94b8e647014d60d2542a336651bf58f"}
{"id": "train_22902", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "reproducibility", "tooling", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "229ce157193cfd82efe56f041f4f43959db010d0"}
{"id": "train_22903", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "tooling", "repo_scale_reasoning", "governance"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "34ff7256f33e8c6f82f43461efc3a03698383dfb"}
{"id": "train_22904", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "auditability", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "855fd810ffe59c02b166509d0e2abd6d5aad6062"}
{"id": "train_22905", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "tests_are_truth", "reproducibility"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ae9b03d45fcd01a7adc1b3675f2b33ab3b0f1c0f"}
{"id": "train_22906", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "repo_scale_reasoning", "tooling", "documentation"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d7c799196d99d8e2d50262adb0993ed203c15623"}
{"id": "train_22907", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "reproducibility", "security_gates"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "25ed7be5b30b06d54aab1e8d99441575517aab9c"}
{"id": "train_22908", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "governance", "repo_scale_reasoning", "auditability"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8010fe499f95803b204053552a4e46e51ceace6c"}
{"id": "train_22909", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "governance", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "08021b03d2768e03e3f4a67b9abdac50997e87cd"}
{"id": "train_22910", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "governance", "evaluation_metrics"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "27beaaeeab855d2314c36ff885a62c77f792b645"}
{"id": "train_22911", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "tests_are_truth", "tooling"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1e9b71f638bcface3a2d9a86925e098b74cf29d7"}
{"id": "train_22912", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "documentation", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f926d59708b3944153b23db9b9e1fba5fdd80552"}
{"id": "train_22913", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "tests_are_truth", "security_gates", "governance"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2bf4646fdf5c2dfcf661c3eeefb2447c2418fe74"}
{"id": "train_22914", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "reproducibility", "documentation", "tests_are_truth"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9b0dd3d0c11f4bbe2da50ed323870408a6fe68a6"}
{"id": "train_22915", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "governance", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "29414ae798e91bfaa371d57ba9e45d1b74fce456"}
{"id": "train_22916", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "ci_integration", "tooling", "documentation"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "008730e7c37f5f670078db504ba882eb474e532d"}
{"id": "train_22917", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "ci_integration", "tooling", "auditability"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "16b51f3aaa197227f5b7f6c3a299ee3bed6b77b1"}
{"id": "train_22918", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "tooling", "reproducibility", "governance"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0ca7d3c069eebe3dc8208a3258b6740c30225e4f"}
{"id": "train_22919", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "cost_latency_tradeoffs", "reproducibility", "auditability"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d7f276ecb05b9a07e360067476ac134feee8f4a4"}
{"id": "train_22920", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "tooling", "ci_integration", "tests_are_truth"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6694ea4272815b87076b19e094c2098cd4d8fc28"}
{"id": "train_22921", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "security_gates", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9615d8e29016e1d8ac9b8a8b33432acac13cb0de"}
{"id": "train_22922", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "tooling", "evaluation_metrics", "governance"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bab0e6eb31cc30d1d27a6ac0b426f8a08e4b0037"}
{"id": "train_22923", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "tooling", "auditability"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ffcbe80116ec4ba122b71b76eb0e99e47cd58f46"}
{"id": "train_22924", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "security_gates", "cost_latency_tradeoffs", "governance"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9c6e0701baf199027c811968dc2d73591ed2941c"}
{"id": "train_22925", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "documentation", "tests_are_truth", "reproducibility"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b15134e997bba863bcde1d82a75b22777469b114"}
{"id": "train_22926", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "auditability", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "105a42a98ed2829cc300d9147d70d9ca653b8765"}
{"id": "train_22927", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "tests_are_truth", "tooling"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3ea22a60edd999adff1a0e9d1618a9eb3e18e517"}
{"id": "train_22928", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "auditability", "governance", "security_gates"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "238839cee8c69e3eb761656987d6ef154e2eb73e"}
{"id": "train_22929", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "evaluation_metrics", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bd080ecae3014e0ac3c0c951b1f8f322a908fcdd"}
{"id": "train_22930", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "auditability", "governance", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5ae10e58249135bde12c7784a1073cfe65090d9d"}
{"id": "train_22931", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "auditability", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2e4ccb2ef5781ce2d68c738846ecc08df5a7d547"}
{"id": "train_22932", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "auditability", "documentation", "tests_are_truth"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a88b2eebd6a430561aec82c19a00b5f38d03156c"}
{"id": "train_22933", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["governance", "evaluation_metrics", "repo_scale_reasoning", "auditability"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "41a60b5b7d5b8f366c07d26af6d8d25a36a05a48"}
{"id": "train_22934", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "evaluation_metrics", "auditability", "tooling"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d0275b71c53e24ad8425d520b4406cddec9a7ec9"}
{"id": "train_22935", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["governance", "tests_are_truth", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2f656ddecc55430e1cb1cc8a8cc42caeeb04f56f"}
{"id": "train_22936", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "security_gates", "auditability", "evaluation_metrics"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3ab9287879cbb1bec898d6ed2c8cc2d532f3647f"}
{"id": "train_22937", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "documentation", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "85c3892a77251ab1a47d6dd7da37253eca1ab715"}
{"id": "train_22938", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "tests_are_truth", "documentation", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "127854640f009b147b84edbf373a38a71b5b2476"}
{"id": "train_22939", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "documentation", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ad07fcdc4d24df4747c917f443b6938aa87467fd"}
{"id": "train_22940", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "ci_integration", "reproducibility", "auditability"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1a2206397cf5ec0489b033bbdc70bfe4feb28d11"}
{"id": "train_22941", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "repo_scale_reasoning", "reproducibility", "tests_are_truth"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6aeee4e439072b707aecd8920e31f3a8fa7dd931"}
{"id": "train_22942", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "repo_scale_reasoning", "ci_integration", "documentation"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6f1b0822b3ab47d89c5a64be3f28479c1ecd923c"}
{"id": "train_22943", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "ci_integration", "documentation", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0112017e6ae9b67a16cce2056db9effb99ae28a1"}
{"id": "train_22944", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "repo_scale_reasoning", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7d3ea98e925864bb10520d499f6d4f9c4f406bf5"}
{"id": "train_22945", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "auditability", "cost_latency_tradeoffs", "governance"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0539173409c183fdf9e16c7fffd41f2eac133fd1"}
{"id": "train_22946", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["governance", "ci_integration", "tooling", "auditability"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "27b53ba36a003b1f5dd380576288dc7cd17771ea"}
{"id": "train_22947", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "governance", "ci_integration", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e7662e4f409034866fcd990e5d192ebf3ad2cfb8"}
{"id": "train_22948", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "auditability", "ci_integration", "tests_are_truth"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1792a90830716f1873cf0297874af708ce6d93de"}
{"id": "train_22949", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "security_gates", "reproducibility", "documentation"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "14be867872d6150878fbc2a00c23c465f99f80fa"}
{"id": "train_22950", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "repo_scale_reasoning", "tooling", "documentation"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e64f74928db65a124cf00640ea9ee30b46f1df3b"}
{"id": "train_22951", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "tests_are_truth", "cost_latency_tradeoffs", "security_gates"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "26587b7ed1814e8a286df1029b887111974e98b3"}
{"id": "train_22952", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "governance", "cost_latency_tradeoffs", "repo_scale_reasoning"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8a90d46d4dd158f7e100962302d6398368e222b6"}
{"id": "train_22953", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "governance", "auditability", "tests_are_truth"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5b9f7253288f843a14f66e9cda8fe154db898b43"}
{"id": "train_22954", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "documentation", "evaluation_metrics", "tooling"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d9357cddafc7283e21cab64e4c0d4e7f2423506c"}
{"id": "train_22955", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "auditability", "security_gates", "tests_are_truth"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "98abfb3c7c2ef2ca908eb0363c76cb5d7b5a4120"}
{"id": "train_22956", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["documentation", "cost_latency_tradeoffs", "governance", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "173fd23a9efe25d66a612ee863f2fc613f2cf10c"}
{"id": "train_22957", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "security_gates", "tests_are_truth", "documentation"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0392945190ca56b272d05d5bbce244f1774e2dda"}
{"id": "train_22958", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "tooling", "ci_integration", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c12b88a227590c48f66ddb4a6a79d98a996f5f78"}
{"id": "train_22959", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "documentation", "ci_integration", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cf9fa46914c2bc2c4be469370a449253497be01c"}
{"id": "train_22960", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "security_gates", "governance", "reproducibility"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "53fd9aba28fd70df0cafa344a6c790bc61834f15"}
{"id": "train_22961", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "repo_scale_reasoning", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "00cfe8408d9784ec1ed1a022dc512faaa8a952e9"}
{"id": "train_22962", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "security_gates", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0173d6c15640edaf3b2fe4b1c96e711a7d8d8126"}
{"id": "train_22963", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "ci_integration", "security_gates", "tests_are_truth"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "691e3f51693564664f1dce49db364981383e48bf"}
{"id": "train_22964", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "auditability", "security_gates", "documentation"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "25cb99ff9b0aa115109194a1c13c9a478a4c3243"}
{"id": "train_22965", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "ci_integration", "tests_are_truth", "reproducibility"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fb57016ea219d7eded6ba98a3d8c6558f0750f89"}
{"id": "train_22966", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "ci_integration", "tooling"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ae9997f129bdf8d9e499c86c40e5b2d0b3296e0b"}
{"id": "train_22967", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "tests_are_truth", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d0a8c71d9c828ee78142dafb26097f29171c24df"}
{"id": "train_22968", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "documentation", "tooling", "security_gates"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "399f81a490de539d95e49e5ff9830070b0045dde"}
{"id": "train_22969", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "tooling", "tests_are_truth", "security_gates"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "73f947114dcc8f45e0f07628bf90b6a09ec81281"}
{"id": "train_22970", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "tooling", "ci_integration", "governance"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c80147d88cd04674969dc7bb8fe53b5f262a36c1"}
{"id": "train_22971", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c6f90a393e4caa8b28cef784bf28b3a2c71e54c6"}
{"id": "train_22972", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "security_gates", "documentation"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2460d23d52eea0d96bb62890af25a1989738858f"}
{"id": "train_22973", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "repo_scale_reasoning", "evaluation_metrics"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "db3ade8898625c94c5189cceb2ef4c1d96997a26"}
{"id": "train_22974", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "security_gates", "repo_scale_reasoning", "evaluation_metrics"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b64393b807db2d5a9c81e2ef2daa7e212a8a8b14"}
{"id": "train_22975", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "tests_are_truth", "security_gates", "repo_scale_reasoning"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "79e19485a73e135e1e9fb6c265fa115e92b459eb"}
{"id": "train_22976", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "security_gates", "evaluation_metrics", "tooling"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "031699eb7a14486a1bd018dcc01e04af61ce0397"}
{"id": "train_22977", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "repo_scale_reasoning", "tooling", "governance"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "88754fd522c230adecbc554d15d0be8abaa124a0"}
{"id": "train_22978", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "repo_scale_reasoning", "documentation", "governance"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bd8ea54936c77bce07b9bd86ee0dc75b93a6a056"}
{"id": "train_22979", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "governance", "auditability", "tests_are_truth"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "03521c9ff5c96d8021e8b4cd6b3088e9a11c0f29"}
{"id": "train_22980", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "reproducibility", "security_gates", "auditability"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2ef461d1fb1d36e485f91fae5f6e7cfc7c069b1a"}
{"id": "train_22981", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "documentation", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9da073c59f24706b549af74c5497c9b23516e041"}
{"id": "train_22982", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "tooling", "auditability"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "71884849d0420b4d8e387fda9afd3ccf69ad29c4"}
{"id": "train_22983", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "repo_scale_reasoning", "auditability", "tests_are_truth"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bce0af83e3011f49d8a0421e7a8ba1cf161b41b3"}
{"id": "train_22984", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "tooling", "ci_integration"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ea3c83a361065eda66986f472aa18b17842cddc3"}
{"id": "train_22985", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["governance", "evaluation_metrics", "tests_are_truth", "ci_integration"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "33b0416fcb421ffdb2b4b78f0d9036925b6bf88f"}
{"id": "train_22986", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "repo_scale_reasoning", "evaluation_metrics", "ci_integration"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "112844fbbf4ac1742c1208cd677a11dc72839ae4"}
{"id": "train_22987", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "ci_integration", "cost_latency_tradeoffs", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "995bcbae683d43387bdadaa99a2bfe234792d021"}
{"id": "train_22988", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["documentation", "auditability", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c9244cb540c0d776a70924bb9490aeca16699ccb"}
{"id": "train_22989", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "tooling", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "02d42b0b401d7350ed78b0571f42ca4dc53c3cfe"}
{"id": "train_22990", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "reproducibility", "evaluation_metrics", "security_gates"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7c4b7d044e4e5388088faa3003bd839d4b9c5856"}
{"id": "train_22991", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "tests_are_truth", "tooling", "auditability"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4f9b775a88bf6ac01394021d7a7f0da09e9e1196"}
{"id": "train_22992", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "auditability", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5ff2bc0dff54e4e9d8d7cc3aa86690f4c951ad58"}
{"id": "train_22993", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "security_gates", "governance"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d5f4bc994779225cb34a9a6b3757769a6530a13d"}
{"id": "train_22994", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "cost_latency_tradeoffs", "repo_scale_reasoning", "reproducibility"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ad91405fb04cfc82058f6f620d2dd8923f5efaa5"}
{"id": "train_22995", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "evaluation_metrics", "security_gates"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "499fe49ef0457fb05f0aa11246c9ec9c4b066a7c"}
{"id": "train_22996", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "security_gates", "governance"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "af581c4fd37c5353943d43c3f25a2aac25b07b0e"}
{"id": "train_22997", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["auditability", "reproducibility", "governance", "ci_integration"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bd7c1adbcbaef5122182303fa63d74e9e28fa08b"}
{"id": "train_22998", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "governance", "repo_scale_reasoning", "tooling"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "aab84702607f96ee9447e7b2ddddc32beb0ceccc"}
{"id": "train_22999", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "repo_scale_reasoning", "governance", "tooling"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1c6f651767365db0bda4887959c5bea19d047f13"}
{"id": "train_23000", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "tests_are_truth", "documentation"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e75face8f95eeebd2a17815399c7cf8177b1bf7e"}
{"id": "train_23001", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "auditability", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a8c888e2a238a8a3c8b770465024deff47f7e5b5"}
{"id": "train_23002", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "governance", "security_gates", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8fa251d4f4b2ebaa62b014c75cbbd8ef04310fe7"}
{"id": "train_23003", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "security_gates", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "58497f74aee9ae46e96c6214ec1b69ac0d4cd2fd"}
{"id": "train_23004", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "governance", "tests_are_truth", "ci_integration"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "905e77cfbfb38e8fa1c145dcd212521e2216e017"}
{"id": "train_23005", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "ci_integration", "tests_are_truth", "governance"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "70917a23d42b94b147137f94e40965fbf2ff3b17"}
{"id": "train_23006", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "governance", "ci_integration"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "91227a25079af73ec1e704bff09ded036c733a09"}
{"id": "train_23007", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "tests_are_truth", "evaluation_metrics", "repo_scale_reasoning"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "089a19431e2105f492a100522dd51fbfe7519f82"}
{"id": "train_23008", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "security_gates", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3c57bac22934e537a6a60cc93b0c9f5855f41cc9"}
{"id": "train_23009", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "security_gates", "repo_scale_reasoning", "auditability"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "87c0f34e08b139fbcfe70defb3a7bc654d26851d"}
{"id": "train_23010", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "governance", "security_gates", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dee4bc85af11249ce86c90a3d6112ab529f65bf0"}
{"id": "train_23011", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["governance", "evaluation_metrics", "documentation", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c4f606f6234e2337a3d8f87f4fd47d0c4476718a"}
{"id": "train_23012", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["auditability", "governance", "evaluation_metrics", "reproducibility"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "10bfdd0209646e15addd52017216d2031c61ee12"}
{"id": "train_23013", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["governance", "evaluation_metrics", "tooling", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3f6e72bb0f2bf72c7f98b65121af341555003ed3"}
{"id": "train_23014", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "evaluation_metrics", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "19e46239d9290b8a3536d1881ca5106da12ac7bb"}
{"id": "train_23015", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "auditability", "security_gates", "repo_scale_reasoning"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f4be40f1d4794b15f05d04eb2f86d900d6fb014e"}
{"id": "train_23016", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "documentation", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "41410ee3ec4f217eac500691593e8f8a4eaf0eed"}
{"id": "train_23017", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "tooling", "evaluation_metrics", "ci_integration"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9e28b5c84bc0df0bd8915f3239fb53e8a20d5d03"}
{"id": "train_23018", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "evaluation_metrics", "security_gates", "auditability"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4dd0264d94027a235ed4c65d222af0f299f99571"}
{"id": "train_23019", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "repo_scale_reasoning", "reproducibility", "evaluation_metrics"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "56aa9f6fc91786642ddcb00d941578180d61fcf1"}
{"id": "train_23020", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "tooling", "auditability", "evaluation_metrics"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fa38dafae1deb68de63804f69b7ed1aeb653fe95"}
{"id": "train_23021", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "tooling", "cost_latency_tradeoffs", "security_gates"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "37ebb0ac38919803b3dfd759aa0da9e5bb9ae7e0"}
{"id": "train_23022", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "tooling", "evaluation_metrics", "governance"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e884ab6f0ce16985269bf8383e9d5f91d5b4eb35"}
{"id": "train_23023", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "tests_are_truth", "auditability", "governance"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0a2997d51689744883a6cee284894b516ec26954"}
{"id": "train_23024", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "repo_scale_reasoning", "evaluation_metrics", "tooling"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "77d4ff7be59b38e1d883411227fc1f19cd574ed1"}
{"id": "train_23025", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "governance", "auditability"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e451a4f3b56baa760b52131d437b6a1bc05bbe36"}
{"id": "train_23026", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "cost_latency_tradeoffs", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7b72e52e0de7ba46f0ba6fafc8a260f3568fc5e5"}
{"id": "train_23027", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "reproducibility", "governance", "security_gates"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "313e38f25d8895bece5286c465f5589f6860b71c"}
{"id": "train_23028", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "security_gates", "evaluation_metrics", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fc9932597b0cdfda9a74234231a93d842045b5cf"}
{"id": "train_23029", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "security_gates", "governance", "auditability"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "642c1a41e2f68a6922050e11bd580974ca40222d"}
{"id": "train_23030", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "governance", "tooling"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "17f85b2c4db95204c653d3fd8fd63b8936525aa3"}
{"id": "train_23031", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "auditability", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "91f709fbdff60230ad5274bb144c0ec59b937613"}
{"id": "train_23032", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "reproducibility", "auditability"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8b6916462c84f7aa582c8669eb4782350f8b0443"}
{"id": "train_23033", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "security_gates", "auditability", "ci_integration"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "17c7cc24ce5fa94723a8832f9e58ae2df50d3bc4"}
{"id": "train_23034", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "tests_are_truth", "security_gates"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fc968f9771736a9199a011fc62893521faf45d1b"}
{"id": "train_23035", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "evaluation_metrics", "ci_integration"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c7bc344e9c7111b0e39b7f9ed69a9ea5ac9c3f47"}
{"id": "train_23036", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "tooling", "evaluation_metrics", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "569f6acbd66fdef33c60d021f930ccd4868e1756"}
{"id": "train_23037", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "documentation", "evaluation_metrics", "governance"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5b9624db2cbbaa4e5b2c96d5c6a3108e86661528"}
{"id": "train_23038", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "repo_scale_reasoning", "evaluation_metrics", "governance"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d5a42477005db143bc9b083c1319ed53ebef360e"}
{"id": "train_23039", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "tooling", "ci_integration", "auditability"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f5c7f12c5cdc28fc7e190a4a8884b1b9abfd46b8"}
{"id": "train_23040", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "auditability", "security_gates", "ci_integration"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "64e8d22feb3a1901500cd00f5ee7acdb10c1b4ae"}
{"id": "train_23041", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "security_gates", "auditability"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "570fe076ccbcf7fd3361adc52a612633bcf0745e"}
{"id": "train_23042", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "governance", "tooling", "security_gates"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6ad5932b3fcc689de893f63b2027ffb6f67b2cbe"}
{"id": "train_23043", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "ci_integration", "reproducibility"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "71eb1252c3f7ae1ca75229b3f3d79ac775c50e43"}
{"id": "train_23044", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "auditability", "documentation", "tooling"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "550b03cc5f8c6cbd954d07833013a144794d1f89"}
{"id": "train_23045", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "evaluation_metrics", "auditability", "ci_integration"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "54578d63e30c78901b960f4a3e19bb6ed32c17f4"}
{"id": "train_23046", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "evaluation_metrics", "auditability", "ci_integration"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "492ba6d8500e00f03fc3dd08024f24f7423abc9a"}
{"id": "train_23047", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "ci_integration", "governance", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b54014240982f5e9af9d671a350c2e0f51503f15"}
{"id": "train_23048", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "security_gates", "tests_are_truth", "documentation"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a68f948b9301ace61336c153a9ad16be9e33ba6d"}
{"id": "train_23049", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "documentation", "security_gates"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6af1ab25b117bfcd75ca1997b0390ecbd62b46a5"}
{"id": "train_23050", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "governance", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "750886e5b3c16dcda42ed6086f7a2a7cc968ba69"}
{"id": "train_23051", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "governance", "tests_are_truth"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d8e84e5d69a67c1d19dd9c4a6326c31f48750948"}
{"id": "train_23052", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["governance", "reproducibility", "security_gates", "auditability"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ae530bab0cfb6ca46bd936307fbdd06d2c1aef05"}
{"id": "train_23053", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "tooling", "tests_are_truth", "ci_integration"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "24983bfc251a6ec44496cf06055852819998e4e2"}
{"id": "train_23054", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "tooling", "tests_are_truth", "ci_integration"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d322738eeec620e2b0935340a0766556197c1d02"}
{"id": "train_23055", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "governance", "ci_integration", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5aae1084e9c90e87a94dc174a449882a7bb86eee"}
{"id": "train_23056", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "auditability", "documentation", "governance"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dde973614d1f4a1e1ec65e368070e64e332e08ce"}
{"id": "train_23057", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "documentation", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7b945fef042b9a5770b26f413d1c2f8d3b18bb4b"}
{"id": "train_23058", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "tooling", "security_gates", "governance"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a65cbbeb53a434fb94e076e5a9fe825c55ef3657"}
{"id": "train_23059", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "tooling", "auditability"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4c8de8c1a677e9251ee7752b62d133ea63e52930"}
{"id": "train_23060", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "tests_are_truth", "ci_integration", "tooling"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1455df681e206a8e5dc5bc9e752d2f03b48f0c84"}
{"id": "train_23061", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "documentation", "cost_latency_tradeoffs", "repo_scale_reasoning"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "82178fed697e5836e7d8db21fcbb0ff04540679b"}
{"id": "train_23062", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "evaluation_metrics", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e0076b0318124b95d786d526138bf4ceeb0195f8"}
{"id": "train_23063", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "governance", "tooling"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1ad645048f724e84627a5324c146f955e2878f78"}
{"id": "train_23064", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "auditability", "ci_integration", "governance"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "010eeee1b4195e0f9f09e072f90d08280db83a1f"}
{"id": "train_23065", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "tooling", "documentation", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "af1e5559c490a3b47c721305cbf71a3a066709ee"}
{"id": "train_23066", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "tooling", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "81a3ed96ab0a0b8d45514e9f9646460f5c8c5695"}
{"id": "train_23067", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "security_gates", "tooling", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3fd2094d92a38c737918753709563a2c97ab0032"}
{"id": "train_23068", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "governance", "repo_scale_reasoning", "reproducibility"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c61c1a861d459c2bc1aa727a2cb16dbf2bd2eb44"}
{"id": "train_23069", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "governance", "tooling", "security_gates"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5bb26c477fedd75e28506ef74d35b5e1e3f4926d"}
{"id": "train_23070", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "auditability", "documentation", "ci_integration"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "21b92a11c7f74c352478e5cac3fb3c9a4d8ea301"}
{"id": "train_23071", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "security_gates", "auditability", "ci_integration"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d7b9b45fbe75c947c8c426b221e25ce683796355"}
{"id": "train_23072", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "auditability", "evaluation_metrics", "ci_integration"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cfe7aff556eadb3f5a3091f6fc4dbd78e210528c"}
{"id": "train_23073", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "evaluation_metrics", "reproducibility", "auditability"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "aa1930c19cca9f0e44f15b2e2aec120af9bb38c4"}
{"id": "train_23074", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "ci_integration", "governance", "auditability"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dcee2f409367b7cbe9a7211981e527d639b797fb"}
{"id": "train_23075", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "auditability", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d5c2be0c741cbff3fa462a8d39ab8bc1e1a24379"}
{"id": "train_23076", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "tooling", "security_gates", "ci_integration"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "535ad9fd4a007bb711c1748289a09d2da9c5d85e"}
{"id": "train_23077", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "governance", "reproducibility", "ci_integration"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ad93776f4be09d9de321bcd68f50ac75991e0d20"}
{"id": "train_23078", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "reproducibility", "documentation", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "414a34330b5681211c388e3beeaf738a2a58b152"}
{"id": "train_23079", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "tooling", "auditability", "ci_integration"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f75796afbd69ad839a014586c13401988eb59122"}
{"id": "train_23080", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["auditability", "tests_are_truth", "tooling", "ci_integration"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8efdc44d13e7e46de2e8c72943d0a63bf98d882b"}
{"id": "train_23081", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "security_gates", "auditability", "tooling"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1e206dcff951e433416ef57887530471ea2eefbb"}
{"id": "train_23082", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "tests_are_truth", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1bd8e1903f1c4c9536ce980a3039513cf305bfe2"}
{"id": "train_23083", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "repo_scale_reasoning", "ci_integration", "documentation"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4dda38b527bb3a65522a495a85a9139c33ac6044"}
{"id": "train_23084", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["auditability", "reproducibility", "documentation", "ci_integration"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "88b8e758678988826d17f3c11ef83c088f2ba5e0"}
{"id": "train_23085", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "auditability", "documentation", "tests_are_truth"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c14e01c1033f52ef67d1f9bb3ad2b84a6c471616"}
{"id": "train_23086", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "auditability", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "beb25849ed0fb5ad2041b7d9f667371c292d3194"}
{"id": "train_23087", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "evaluation_metrics", "tooling", "ci_integration"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cf27421abfdaaaec5b228d36dca7d2c3a4a6c06b"}
{"id": "train_23088", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "governance", "evaluation_metrics", "repo_scale_reasoning"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d0ea7cc9b972fede0e368386eb07c9207522c242"}
{"id": "train_23089", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "auditability", "reproducibility", "ci_integration"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f417b35c25328b9f4857d2e5cae0f48498903570"}
{"id": "train_23090", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "auditability", "cost_latency_tradeoffs", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d05c6ec4ce06c90d8e49528f1f983d53fd961d1f"}
{"id": "train_23091", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "security_gates", "ci_integration"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ed4b91f70bf539922d25f3048b1fd2e5d493dfac"}
{"id": "train_23092", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "security_gates", "ci_integration", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "aef00630c8239c4af69078fffa7e8a74129a330f"}
{"id": "train_23093", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "cost_latency_tradeoffs", "documentation", "reproducibility"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6ed0551779f5236100d8c0e05128699eae668f8b"}
{"id": "train_23094", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "governance", "security_gates", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e0b1cafddcad3ef915d2dd99265307c529f97158"}
{"id": "train_23095", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "evaluation_metrics", "ci_integration", "security_gates"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d34ad907b4e3a460d29ad26e1b72d1c0e3ed7b79"}
{"id": "train_23096", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "reproducibility", "tests_are_truth", "documentation"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "743c3b3ae1be84e14a930963c964e85a2e1ffcc3"}
{"id": "train_23097", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "evaluation_metrics", "tests_are_truth", "auditability"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d44a6604c82c258b09fddf2940a4ad5209a2f680"}
{"id": "train_23098", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "evaluation_metrics", "auditability", "tooling"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6cf2355ef8a20943b546e92e21b493f66a2063b0"}
{"id": "train_23099", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "ci_integration", "reproducibility", "tooling"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4f91ecb039fa040a5d167fcc455180357800ab84"}
{"id": "train_23100", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "auditability", "cost_latency_tradeoffs", "security_gates"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "24cbe6e22ef86e0d07bb00c2f0d60e126330372a"}
{"id": "train_23101", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "security_gates", "ci_integration", "documentation"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8a43ffebfc239361cacbe2d6060c66a0c44db2e2"}
{"id": "train_23102", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "documentation", "auditability", "ci_integration"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0ba0bc4afdf049f337cc149ab285ed4cc009d558"}
{"id": "train_23103", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "repo_scale_reasoning", "ci_integration", "auditability"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f0b86fbe210cd60b5dfa7c4e9fc8e982f8355ab6"}
{"id": "train_23104", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "reproducibility", "governance", "ci_integration"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "54fa13c0d817d38ba5c58eb152a9459bcdbe4073"}
{"id": "train_23105", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "tooling", "evaluation_metrics", "auditability"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b3976a2d35c6949590a3714eae958de8ed20412d"}
{"id": "train_23106", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "evaluation_metrics", "auditability", "security_gates"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "381355bc057458fe524e37c2c5b19a73b8f753df"}
{"id": "train_23107", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "security_gates", "reproducibility"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2ebf110f3a4558dbdaef4517a0e8b0087182cc46"}
{"id": "train_23108", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "governance", "auditability", "tests_are_truth"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c30b11a2584961f0fb8ac9a737c7a604c064ed7c"}
{"id": "train_23109", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "tooling", "security_gates", "ci_integration"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1630c041c972a4dfead99c2fc04f7f076892b931"}
{"id": "train_23110", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "evaluation_metrics", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6f520e364f5d68ac498fabe6552df1954381faeb"}
{"id": "train_23111", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "ci_integration", "security_gates", "auditability"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3e79dac0a6946fb66030be12295e3fde044cea1a"}
{"id": "train_23112", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "reproducibility", "governance"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "36ccb46ef56c2b946b910677b43cf890c54d6444"}
{"id": "train_23113", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "tooling", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e003076d1d76660cf7bf6d7c78ae896c3bd2066c"}
{"id": "train_23114", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "reproducibility", "documentation", "ci_integration"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "aa89f599f2cbcdf0226cf81de46ddd76a5ae7b4f"}
{"id": "train_23115", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "documentation", "ci_integration", "governance"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "214313c47b66218f466ec151cd7714113228181f"}
{"id": "train_23116", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "reproducibility", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7e5e1353ad147f949bd5f0db92046829aecbb99b"}
{"id": "train_23117", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "repo_scale_reasoning", "tooling", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "03643bf99d584459163b43f4f2261f2da29f5d76"}
{"id": "train_23118", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "security_gates", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f30915c15bebb00882c362b69e16e0d36d493fdb"}
{"id": "train_23119", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "documentation", "evaluation_metrics", "tooling"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8668cf27183ab8ab969b901045435e5f476d3f1f"}
{"id": "train_23120", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "evaluation_metrics", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "62e32ee3d67e4e3ce5a37ad683d23a0085d82d5c"}
{"id": "train_23121", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "ci_integration", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "51a62dd8adfafeaa6b01689a89559245023c6f6d"}
{"id": "train_23122", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "governance", "ci_integration", "documentation"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d86d38ef2992abaa6e78359d0475089870cecf97"}
{"id": "train_23123", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "tooling", "ci_integration", "auditability"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0211781853857bff673fb6810766afd699dddc68"}
{"id": "train_23124", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "security_gates", "reproducibility", "tests_are_truth"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "41fae4108a70e3237d16fb0cb5ec82b2ac17d7dd"}
{"id": "train_23125", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "reproducibility", "cost_latency_tradeoffs", "governance"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "92c7a2b7d39761d9e83d92d6789a940069b2a64b"}
{"id": "train_23126", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "repo_scale_reasoning", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "20d10197250ae320a558c3dd8c0a70009e666ab3"}
{"id": "train_23127", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "reproducibility", "documentation", "governance"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cff9c66a6c0e6d88f4b9259225202471708295ad"}
{"id": "train_23128", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "governance", "tooling", "evaluation_metrics"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3b41522eed59c4d83111f0da4d152b3a3e1f6f7c"}
{"id": "train_23129", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "documentation", "cost_latency_tradeoffs", "repo_scale_reasoning"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e451a4f3b56baa760b52131d437b6a1bc05bbe36"}
{"id": "train_23130", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "governance", "ci_integration", "reproducibility"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "549db2b2d1c5ea6372ab52672d077a048ed9303a"}
{"id": "train_23131", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "reproducibility", "evaluation_metrics", "tests_are_truth"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d887e036460bc3c0a067043e66c8357f4ca4a0a0"}
{"id": "train_23132", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "tests_are_truth", "auditability", "governance"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "757cf394289043792c3830f0aee58d717299bdcc"}
{"id": "train_23133", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "governance", "tests_are_truth", "security_gates"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2f75b9ef52f386c951da4d2369c5c4f4c2540f21"}
{"id": "train_23134", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "auditability", "reproducibility", "tests_are_truth"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2bf7cb2087eee20060ce705440c6374d715a31c4"}
{"id": "train_23135", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "reproducibility", "evaluation_metrics"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "eb48a5a771e86f4e725b495f468fae4cfbfa1e7d"}
{"id": "train_23136", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "reproducibility", "tooling", "ci_integration"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "44f5e80d0db7435c849aed0e151e5d07f186d92c"}
{"id": "train_23137", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "reproducibility", "governance", "tests_are_truth"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "919f8e8a825204f9e680c9382eb80052719e4475"}
{"id": "train_23138", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "ci_integration", "tests_are_truth", "auditability"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2463cd4d099335aaeebead94fc412af39c8a881d"}
{"id": "train_23139", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "documentation", "evaluation_metrics", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "225bf45495973ff22f198549d6b06ec7bd201f51"}
{"id": "train_23140", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "tooling", "ci_integration"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8bcabc9c52af2adbb2ee801df3f0adb4a83af589"}
{"id": "train_23141", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["governance", "reproducibility", "tooling", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1d90c1ac4e4d984dcd374abeaed5d5ad612cdf52"}
{"id": "train_23142", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "auditability", "security_gates", "documentation"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2be96933116794d2617ac1ba8b9d58768fc47d65"}
{"id": "train_23143", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "security_gates", "tooling", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "74a9f4937e1c8a663d019c20b46c6a7fda35b96e"}
{"id": "train_23144", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "evaluation_metrics", "ci_integration", "tests_are_truth"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2a265d862bb6f00b5e8e0d5208c6087785b0e12f"}
{"id": "train_23145", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "auditability", "tooling", "tests_are_truth"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "37bff38f43bfe58403adce4f8d2a6cda06b7c1e2"}
{"id": "train_23146", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "cost_latency_tradeoffs", "governance", "security_gates"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5fa7aac3248804d1de77b2b8b6b2b4e8df6bd57d"}
{"id": "train_23147", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "tooling", "tests_are_truth", "ci_integration"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1348e4619358cc908e724c23b45487e75f80faa2"}
{"id": "train_23148", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "auditability", "documentation", "tests_are_truth"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7cc87ae79683495170804956536e7c5b2752ade5"}
{"id": "train_23149", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "evaluation_metrics", "ci_integration", "security_gates"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3acd2f1743fc924339c3fd3e8bf90aebee90967e"}
{"id": "train_23150", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "documentation", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "54b47341dcd47626ee9e3dea2d5907eac13d8f02"}
{"id": "train_23151", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "auditability", "governance", "security_gates"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2777a5a808071f87228e5c3622c11589346f78d2"}
{"id": "train_23152", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "repo_scale_reasoning", "documentation", "tests_are_truth"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1533cee1db8a38c6d5e3aadd2ed4d6f346a6dfba"}
{"id": "train_23153", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "evaluation_metrics", "governance", "security_gates"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "09d2d0194f6c36d1b5604d4a35ff45f00cf19228"}
{"id": "train_23154", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "security_gates", "tests_are_truth", "reproducibility"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "252f8276b29ee6ce38ee7efcd4856a3f66478e10"}
{"id": "train_23155", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["governance", "documentation", "evaluation_metrics", "tests_are_truth"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b37e311984abe34954b5188f8bb6e54fd8b7a054"}
{"id": "train_23156", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "reproducibility", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8d5c9dfd83dd398057f6dc7af027bf4e72a77524"}
{"id": "train_23157", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "auditability", "governance"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ae8026f85da246e44ec83d458abb6e4f9b0c2821"}
{"id": "train_23158", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "governance", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0bb1cb3b4d3af417ab7d74585ce04dc1b0f38d19"}
{"id": "train_23159", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "documentation", "evaluation_metrics", "security_gates"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "09d6740acbdc41e0b4308696753e103e59e1b1f4"}
{"id": "train_23160", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "security_gates", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "39ff962bd6eafeaf0d7633120bd05289e0a16add"}
{"id": "train_23161", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "auditability", "ci_integration", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0d8deba48096832be99b16ec86d3eccc61789732"}
{"id": "train_23162", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "repo_scale_reasoning", "governance", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dc7c872c0b6df6b3d3857027c5080fb325df8bbf"}
{"id": "train_23163", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "documentation", "evaluation_metrics", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ef8c3e27501c58a1916e70ff498fdf9f54c0c912"}
{"id": "train_23164", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "documentation", "governance", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cbd60889e02e7bb8b38b397b3128806a3bf9dc0c"}
{"id": "train_23165", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "repo_scale_reasoning", "tests_are_truth", "ci_integration"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d2e5924673b2628d1cd839dc14300b1db8582d96"}
{"id": "train_23166", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "evaluation_metrics", "reproducibility"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0a7ec494900a1c6848cdbcea99f254cab91276c1"}
{"id": "train_23167", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "repo_scale_reasoning", "auditability", "reproducibility"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6584e2eb6874dbc1497cd5ac777202a76d0adf76"}
{"id": "train_23168", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "ci_integration", "evaluation_metrics", "reproducibility"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5ab5ff8c8094cc8fcd7f7adfd7c9b6eafb41dcb5"}
{"id": "train_23169", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "tests_are_truth", "evaluation_metrics", "ci_integration"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dd0d942d227ed4c050a130d9e6411659f4d9f01a"}
{"id": "train_23170", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "reproducibility", "tooling", "documentation"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2000560a3532c59d9e88e3fd110df792183f5237"}
{"id": "train_23171", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["auditability", "evaluation_metrics", "security_gates", "documentation"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a63521cfa1c4a3c72f6352814420c7fa7c512fd8"}
{"id": "train_23172", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "evaluation_metrics", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "76aa184af11cfe65f0d0c05d1921f86b14441e23"}
{"id": "train_23173", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "documentation", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "56971a0893d575866df557ef452ba2c6dd9e29ac"}
{"id": "train_23174", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "tests_are_truth", "documentation", "security_gates"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f1048071e711606f5afd5361d710ed62f1a30e2f"}
{"id": "train_23175", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["governance", "tooling", "documentation", "ci_integration"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "882f83b546a7264dafd9b26949c5ff6070f072ec"}
{"id": "train_23176", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "governance", "tooling", "security_gates"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f103eea38dce3e1ffb4b6b5218329b089d641d88"}
{"id": "train_23177", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "governance", "tooling"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e8094b5902d5bda6aad4058234551aa50065b55b"}
{"id": "train_23178", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "reproducibility", "tooling", "tests_are_truth"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "44cd33a06426f971bfaf580c57a932ce7b564a7f"}
{"id": "train_23179", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["auditability", "reproducibility", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9f657141532ecdd7d29c17a340eb7b27bbbd0552"}
{"id": "train_23180", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "reproducibility", "governance", "ci_integration"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "07088dda478536294935a041f458d74bd6d7faff"}
{"id": "train_23181", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "auditability", "tooling", "documentation"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ca7df84a2b08787fa18c7a927fa7624c8140034d"}
{"id": "train_23182", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "security_gates", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "813619a79aaf18ff0d10e521e5f19965c5b77e44"}
{"id": "train_23183", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "tests_are_truth", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d5a628d00cdc45c13e164868614aa35d3acaac6f"}
{"id": "train_23184", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "documentation", "tests_are_truth", "tooling"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2f4ec9cf6c65e26359f2be4fd80b5d659c5cdd53"}
{"id": "train_23185", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "reproducibility", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "da40e6d646845b3b9bd36e1b3dc23e567dbbfbe9"}
{"id": "train_23186", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "security_gates", "reproducibility", "ci_integration"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d22797340422bb1d0d7a2dc2e9e62d62810bdd88"}
{"id": "train_23187", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "documentation", "tooling", "ci_integration"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "59d974c5117a9486a7345296f7759fc19a6c32ed"}
{"id": "train_23188", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "security_gates", "ci_integration", "tooling"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ca4f2386e34b94c8d394c000f9b2afa8d2316bb3"}
{"id": "train_23189", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "repo_scale_reasoning", "tests_are_truth", "tooling"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d24ddbd475cb7bf8b3e680269dc96028463de666"}
{"id": "train_23190", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "governance", "security_gates"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7f096ff6f817cb45cb904ec5d71cb40eecb6a261"}
{"id": "train_23191", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "reproducibility", "auditability", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "688f1968888b00e5f0a72ee413721db650527232"}
{"id": "train_23192", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "governance", "documentation", "ci_integration"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e50c942bc705d01435dbe78b20fa917f8971768c"}
{"id": "train_23193", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "evaluation_metrics", "security_gates", "tests_are_truth"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4532bcd9c8fa71ca73d31fed3ba4fa85403f9025"}
{"id": "train_23194", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["documentation", "repo_scale_reasoning", "ci_integration", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "61e916c6b463deaa45e535c50f67dad6c16a5479"}
{"id": "train_23195", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "documentation", "ci_integration", "tooling"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "47b39ab29f9afcec2d53f12282ea853f116f5912"}
{"id": "train_23196", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "ci_integration", "auditability", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "af402df6e37751a7f51badc131a52cc73f2db60b"}
{"id": "train_23197", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "security_gates", "auditability", "tooling"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "787129152f7be2c0beb93db10e5b30a32295163c"}
{"id": "train_23198", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "ci_integration", "tooling", "tests_are_truth"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "70438cee26a353aa8d8db3262c0e7012d8e81198"}
{"id": "train_23199", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "reproducibility", "evaluation_metrics", "auditability"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8c619adae3c1cd8ba8adf5051c97df42e025c3be"}
{"id": "train_23200", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "reproducibility", "governance", "documentation"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a852e8c403e2a0d9ee721ad6c6714e1a40af01e0"}
{"id": "train_23201", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["governance", "auditability", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5c01f47e538459b022ce1d58667af79257adfcfa"}
{"id": "train_23202", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "governance", "ci_integration", "documentation"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "739e8ec84b7eb3dedf5e164c2d500f7fc059650b"}
{"id": "train_23203", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "tests_are_truth", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "aadd17fdc36d49c4f794117dae6f3019b39df5fe"}
{"id": "train_23204", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "reproducibility", "evaluation_metrics", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1563357e66c2480dae2522844ccaed364f69c1d0"}
{"id": "train_23205", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "ci_integration", "auditability"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "827d313e9acfbf968edc36b24dfd2406931b91b4"}
{"id": "train_23206", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "evaluation_metrics", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "39a138030ee4ac0d3464c15849128c20f830eade"}
{"id": "train_23207", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "evaluation_metrics", "tooling", "ci_integration"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "62f0bcad36891d67012cea4a580bb2f157419ca0"}
{"id": "train_23208", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "tooling", "evaluation_metrics", "reproducibility"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fc0c38d6ec6cfb350cb8ee518dad0be203c9c7ae"}
{"id": "train_23209", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "evaluation_metrics", "governance", "reproducibility"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e3e545e17779e715ac43b71dce543dbcbf8d17bf"}
{"id": "train_23210", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "ci_integration", "tooling", "documentation"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "48ce973c9abf1b4d121609ff4de08fa8e2dd07ae"}
{"id": "train_23211", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "evaluation_metrics", "tests_are_truth", "documentation"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e2afc4e085baa518f62ea2f7d111b7498cb30531"}
{"id": "train_23212", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "repo_scale_reasoning", "ci_integration", "governance"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "96b3048b6ac0ef9f491310ec8b0ef70ce72a81b3"}
{"id": "train_23213", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "security_gates", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a664c64d73624c367bfda1d141dedb47163f6e7f"}
{"id": "train_23214", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["governance", "evaluation_metrics", "auditability", "reproducibility"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9dcf720f523e9a761e0c813d5a9ab087e78158ae"}
{"id": "train_23215", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "repo_scale_reasoning", "evaluation_metrics", "auditability"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4341a5c69b468576b0009345bb1a0427a3799957"}
{"id": "train_23216", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "auditability", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ffda027be9afaf3bf6725024c8015d80994b7bf4"}
{"id": "train_23217", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "tests_are_truth", "reproducibility"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6061700c7b21f59048ae32530f6b0406b66322af"}
{"id": "train_23218", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "tests_are_truth", "auditability"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "045b4a07f3131d39038e86fee32a79a2067b8709"}
{"id": "train_23219", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "security_gates", "auditability", "governance"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dd7d941bd9ee9011a3153f7917bdb3a1d6068194"}
{"id": "train_23220", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "reproducibility", "cost_latency_tradeoffs", "security_gates"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "eaf57b6f71597ae486643557838fa24fa60c29d5"}
{"id": "train_23221", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "tests_are_truth", "tooling", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c45071c31d39f7bec084a48276b42fa695636800"}
{"id": "train_23222", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "security_gates", "repo_scale_reasoning", "documentation"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "26bb7349ab6e0f3b7872cad5869d1565c14454d7"}
{"id": "train_23223", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "security_gates", "tooling", "ci_integration"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d37c7c8ae68b5905da259d53f6a8e9464db2d6e7"}
{"id": "train_23224", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "documentation", "tests_are_truth", "auditability"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "13ba5ea0e71bb10dc912f27d17c3e6700179f0e6"}
{"id": "train_23225", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "security_gates", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cb824a662d882fb454fb248e5408a8a21b8c955f"}
{"id": "train_23226", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "auditability", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "500ebc9e6f05abab3c870640b6b09f9f761c2e36"}
{"id": "train_23227", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "governance", "security_gates", "evaluation_metrics"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6e46a25b0445b655c98f2f3590e85a5a6b805102"}
{"id": "train_23228", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "ci_integration", "evaluation_metrics", "governance"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "691dd951107cb7fc9e617ab60d67df3466e54451"}
{"id": "train_23229", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "documentation", "auditability", "reproducibility"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f93b3ae0c9a3fb276209830924d3f441d769b1c0"}
{"id": "train_23230", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "reproducibility", "evaluation_metrics", "ci_integration"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b8e585307df0095c6e25ec5c47b1b7320a4d8161"}
{"id": "train_23231", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "documentation", "security_gates", "governance"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6b8f85dae81fa224cd9e36fb26c9dbc3125c3040"}
{"id": "train_23232", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "security_gates", "governance", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4e24030bb0e7cd624d3cf8091ea0101e906efd01"}
{"id": "train_23233", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "repo_scale_reasoning", "evaluation_metrics", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a1269276b608f2d9fe61896fc21d1dd49f9cd9c0"}
{"id": "train_23234", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "ci_integration", "evaluation_metrics", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "70fffef3fe1e848984682ee96998bbfcc46e4539"}
{"id": "train_23235", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "repo_scale_reasoning", "security_gates", "tooling"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "67275c930ba3302f172c84b7951893b769959372"}
{"id": "train_23236", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "tooling", "ci_integration", "documentation"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "48834cac8ac3c309812c0c7c86a509cd46909108"}
{"id": "train_23237", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "security_gates", "tooling"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fcdad25a14898069728c1a07b299d9ec1a57de72"}
{"id": "train_23238", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "documentation", "security_gates", "reproducibility"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "716048173e95cdc03eec57f2edeacf7837335504"}
{"id": "train_23239", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "governance", "tests_are_truth", "tooling"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "abb25833c194b5e7a3ba846f654d8ae6579e5f1c"}
{"id": "train_23240", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "documentation", "tooling", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a3a1b3dae635a45ea975f870d871f5eeb0407fed"}
{"id": "train_23241", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["auditability", "security_gates", "reproducibility", "tests_are_truth"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "82ba36cc9b301f49a28186be565a8b52d50fb76f"}
{"id": "train_23242", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "security_gates", "repo_scale_reasoning", "reproducibility"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5eb28894b260372666bdc12703d51becaffccec7"}
{"id": "train_23243", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "ci_integration", "auditability"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "37926d4ed4915e942f589f4097ba7fda892274fb"}
{"id": "train_23244", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "tests_are_truth", "tooling", "reproducibility"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8dea5d220a0ffa467db3a811aeae6cdee0b9631b"}
{"id": "train_23245", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "tests_are_truth", "reproducibility", "documentation"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f19e40f319658e7f54d691bbd4292f85ce648f93"}
{"id": "train_23246", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "tooling", "governance", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "98f9dc07c9e5f05b1770734959b0dff80d5c49ee"}
{"id": "train_23247", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "governance", "auditability", "repo_scale_reasoning"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6e8994b925c191aa94fc9c158a576d902d9c6e73"}
{"id": "train_23248", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "auditability", "tooling"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a03c6a7c16b28ec4c32d01185375a3c589035e89"}
{"id": "train_23249", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "tooling", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "963702baf8d86ab6fb3adea400b82ef0cc2204bf"}
{"id": "train_23250", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "documentation", "auditability"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c0098255836f3817f179f1a99f625bd5961b7d9c"}
{"id": "train_23251", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["governance", "documentation", "security_gates", "ci_integration"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "08c3a1332ee40457eb6360d0efdd662b05f1b2e3"}
{"id": "train_23252", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "tooling", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d341a3a923cc2d0a2ed74a35d8a4a6fced85ae47"}
{"id": "train_23253", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "ci_integration", "tests_are_truth", "security_gates"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "918f2ecf88bace9ecc9a904aa73afc6fc0e3c2da"}
{"id": "train_23254", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "governance", "auditability"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2efb5ff23bc25cdfbaa91d21b504b6bdbe602436"}
{"id": "train_23255", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "ci_integration", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2965f1c412d605d24b2e59f40a3d4de74adb6097"}
{"id": "train_23256", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "security_gates", "ci_integration", "auditability"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5ad213088ebfa4db02c4b9df9d1c1397abd0a7b7"}
{"id": "train_23257", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "cost_latency_tradeoffs", "security_gates", "tests_are_truth"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ee12e4bc3e46478db62f23e3a8251a3bf9ac4fd4"}
{"id": "train_23258", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "ci_integration", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9b97134354e80d1227e7161fff7f09b429a8c71c"}
{"id": "train_23259", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "reproducibility", "ci_integration", "documentation"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "58f4419d58b3045e5593a2578a96a6e67f1e1078"}
{"id": "train_23260", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "reproducibility", "auditability"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8adf799b4f54b181061f654e6dd854c6daa9f897"}
{"id": "train_23261", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "auditability", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d964a2d5b9ae9e722ae92c063f189f7bd0d433fb"}
{"id": "train_23262", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "governance", "security_gates", "documentation"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0f1f90492109a6b798dcab8acc538b107d28e592"}
{"id": "train_23263", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "governance", "evaluation_metrics", "ci_integration"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1bcd784b9dba6574ae341de9e0a38b6bd87de2fb"}
{"id": "train_23264", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "security_gates", "auditability", "documentation"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a5b47fe764be3d653dedad3ced26910534e7cfd5"}
{"id": "train_23265", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "evaluation_metrics", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cda31a7e6b142bf286fc9840472132f69931f7ac"}
{"id": "train_23266", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "security_gates", "reproducibility"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "82ed5e2ad9ec30e148088d7876c3b09bc3a21237"}
{"id": "train_23267", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "ci_integration", "governance", "evaluation_metrics"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "325094c25160b6c9dc72b34d1c3d4f9851870c9f"}
{"id": "train_23268", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "tooling", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dbea67cc4cacb3b137fc4b3d5d8fe093ab2cbbf4"}
{"id": "train_23269", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["governance", "tooling", "evaluation_metrics", "auditability"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2d35e8b9f0b3922236c08247e96039cf19665108"}
{"id": "train_23270", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "reproducibility", "auditability", "tests_are_truth"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0289e9f97c1e9e5f98c6b90c76401acfc51d9471"}
{"id": "train_23271", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["documentation", "reproducibility", "security_gates", "tooling"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c4b2613008140dcce46fe11c9e9935bb02e87bf0"}
{"id": "train_23272", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "ci_integration", "repo_scale_reasoning", "tooling"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0f65f4e028d5d5a5b5f4a5d278c0106b2e728b2b"}
{"id": "train_23273", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "documentation", "tests_are_truth", "tooling"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8d576a6f393304540ed05ab2a93494b25d38e80a"}
{"id": "train_23274", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "repo_scale_reasoning", "evaluation_metrics", "tooling"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "02b2ee6a73873f6079cf1c50af66efbbca72c3fe"}
{"id": "train_23275", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "governance", "repo_scale_reasoning", "reproducibility"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "933290ce79d40ab0d1dd07f30f9a91d07867499c"}
{"id": "train_23276", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "auditability", "tests_are_truth", "reproducibility"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e1d50d89ab4b7b529c2cbf562b2517d585969033"}
{"id": "train_23277", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "security_gates", "evaluation_metrics"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ba59e24cc6cbffc08d653b91303b94866cbf3fcc"}
{"id": "train_23278", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "documentation", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0a900d217452a73ba952efcfaa678cd0a7f04b09"}
{"id": "train_23279", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "reproducibility", "cost_latency_tradeoffs", "governance"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ab7ffacff5e2f307fa820066718a7a364f606d3c"}
{"id": "train_23280", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "reproducibility", "tooling", "auditability"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "aed8eb0cfaa56de2938213e3cc554552c8666796"}
{"id": "train_23281", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "governance", "evaluation_metrics", "security_gates"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "aeac8e982f9737db89781ef1fe36cdc0afacc692"}
{"id": "train_23282", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "reproducibility", "governance", "tooling"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dc9b36e8e8a573df7e585a4c1d27951eb47811bc"}
{"id": "train_23283", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "ci_integration", "reproducibility"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "44f9ccfad59a967048dfe6354926c29dcdf797f6"}
{"id": "train_23284", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "governance", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "14b2e488b0c2388d052a9ec2ff9c531a1a774997"}
{"id": "train_23285", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "ci_integration", "auditability", "governance"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "533d4e0745ecd454008377e2b4886665df532714"}
{"id": "train_23286", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "governance", "tests_are_truth", "auditability"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "edc6cfcc4cc8ac7a6fb43bf90a61fa0cdf04e289"}
{"id": "train_23287", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "evaluation_metrics", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "97d0020dbbf187e0a317782b411d91b263998092"}
{"id": "train_23288", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "documentation", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7389093a43570315b1b73f63c5dcadb1803579c3"}
{"id": "train_23289", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "governance", "tests_are_truth", "security_gates"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "331678bb29d96653859a0b5e5561bb54c8d84fed"}
{"id": "train_23290", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "security_gates", "governance", "ci_integration"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "68100dafaf01a381b498dfd620a1882b4d200786"}
{"id": "train_23291", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "tests_are_truth", "tooling", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "974eb9bb0e474efa52d7ba6bd689913fe242015b"}
{"id": "train_23292", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["governance", "tooling", "evaluation_metrics", "reproducibility"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "accee427021f07bf7b75a9519e1eef8622c0eec0"}
{"id": "train_23293", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["auditability", "governance", "evaluation_metrics", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "28679df8ec9a08d2b8f9950ce64d72e98c8b5641"}
{"id": "train_23294", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "ci_integration", "evaluation_metrics", "tests_are_truth"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d0be4e3c9be0f5c0afed56c677792887af132732"}
{"id": "train_23295", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "repo_scale_reasoning", "tests_are_truth", "governance"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "63667a6a749b967300c96f5e2b905536fa8a977b"}
{"id": "train_23296", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "reproducibility", "tooling", "documentation"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "126c8d20dd6a5918a97bb7f89fda0eebbb7de216"}
{"id": "train_23297", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "security_gates", "tests_are_truth", "documentation"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7f4ea686677563d7f401b88ad483cf790761e3c0"}
{"id": "train_23298", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "tests_are_truth", "security_gates", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fe275779e96d18f7e87152604f6324d1a911b164"}
{"id": "train_23299", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "documentation", "auditability", "governance"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a8c5b9d0e49023e5e8c19307847b7e2a13814d5b"}
{"id": "train_23300", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "auditability", "security_gates"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "54df691a62cbe40d4154263fd0aaba9e7a5e78ca"}
{"id": "train_23301", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "auditability", "reproducibility", "security_gates"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7c86954a449adfcc2e237bc607d4b048867cc04f"}
{"id": "train_23302", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "documentation", "auditability"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3d44ed6da6f923876af53cd1d22197fcc0c56b38"}
{"id": "train_23303", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "tooling", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "871084cfc50a12342eb9ffea2591514c0cc42082"}
{"id": "train_23304", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "reproducibility", "ci_integration", "auditability"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "afd736b255b5d3a65ad5444160abb1500914624e"}
{"id": "train_23305", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "governance", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f7668d5e0604efa14053d01c835b8a1ce9b117a3"}
{"id": "train_23306", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "tooling", "security_gates", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d8514a0678be3cd517266949e266b34850f42c1b"}
{"id": "train_23307", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "security_gates", "ci_integration", "tooling"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "947e71047ca350e691d2f88d0ac3305def9916bc"}
{"id": "train_23308", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "reproducibility", "auditability", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dd180f82d84f48d42e1258f17266f55ac45c0067"}
{"id": "train_23309", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "tests_are_truth", "evaluation_metrics", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fb1721ccb99e1bdf0ece333803962b2f4ce57dd8"}
{"id": "train_23310", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "security_gates", "governance", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "873c59c57962e446afac7f64bc2d515888c11d01"}
{"id": "train_23311", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "governance", "tests_are_truth"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ec17b853fee00c4f9e51d5be8ca1d5b6c83d3940"}
{"id": "train_23312", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "auditability", "evaluation_metrics", "security_gates"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "509acf19c89aa304ca560ba8f44ff4663848223d"}
{"id": "train_23313", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "reproducibility", "documentation"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2a4465c916919678dae2a485b680433883676099"}
{"id": "train_23314", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "ci_integration", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0d4cb85cabe58bd8c179b0c45c4e91e9914086c0"}
{"id": "train_23315", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "auditability", "documentation"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d57ef54b77b9abc2bda82bc02be12dd3c4e106a8"}
{"id": "train_23316", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "security_gates", "ci_integration", "evaluation_metrics"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6cf2f3cc7e1d7521b5dd1e473fc260aa8ab8e48d"}
{"id": "train_23317", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["governance", "ci_integration", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "905c3276635e2e0505727a83c0b1395cb98a9177"}
{"id": "train_23318", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "tooling", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "711bee2df4646b1ec1b57bc665d30a214444cb55"}
{"id": "train_23319", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "reproducibility", "evaluation_metrics", "security_gates"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8e411caaf4983aeb56b67262a165452f93bc3257"}
{"id": "train_23320", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "governance", "tooling", "documentation"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "447da26eb53a51420fe59bba4e22b5a3474c126a"}
{"id": "train_23321", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "tests_are_truth", "auditability", "ci_integration"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "28421c6b5e1b5083ac88f90050578ab0b7dd46f2"}
{"id": "train_23322", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "evaluation_metrics", "security_gates", "ci_integration"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "33f80798321daf1dafa9d753b7e7c422866b5a7a"}
{"id": "train_23323", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "tests_are_truth", "evaluation_metrics", "reproducibility"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8180172c783d4bf8d5254abbcfd1b82d3951ca03"}
{"id": "train_23324", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "cost_latency_tradeoffs", "ci_integration", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "74a92bbe1c3fe994807ec0572076fb15eb875440"}
{"id": "train_23325", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "reproducibility", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b3dd38abe69a2da678cb01e78e816fe2248b4a1b"}
{"id": "train_23326", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "auditability", "evaluation_metrics", "security_gates"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8ff2adb85aa0895e4b46bef70ac5d991a0bc234e"}
{"id": "train_23327", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "evaluation_metrics", "security_gates", "ci_integration"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b8494c0b9a04cc1e6f3b595133e46182b8b18724"}
{"id": "train_23328", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "reproducibility", "tests_are_truth", "governance"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5333b0242e102cab860e302bcf190adf6f797f2d"}
{"id": "train_23329", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "repo_scale_reasoning", "security_gates", "evaluation_metrics"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "73ba525496d7f5e7d21da1092dc4e0c5f33746e1"}
{"id": "train_23330", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "auditability", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "84218be8befabb0881644a0642764a609b2224d9"}
{"id": "train_23331", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["governance", "security_gates", "tooling", "ci_integration"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3bfd923aa1eec938fe42a9022e5975c592ba1cda"}
{"id": "train_23332", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "security_gates", "governance", "auditability"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0b841cfc7d03819b30ff70aeff5a6c7ee8ae0249"}
{"id": "train_23333", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "tooling", "ci_integration", "governance"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d5002a77096709517dd028501473baa241239038"}
{"id": "train_23334", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "governance", "repo_scale_reasoning", "auditability"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a4f826c99e851c51112aa8ae83d81b0c5cd695c0"}
{"id": "train_23335", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "reproducibility", "ci_integration", "tooling"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "139f5bff4a5875e47d8261d8b9ad22eb7b557e98"}
{"id": "train_23336", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "tests_are_truth", "governance", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "44ff8e52ccd6861d5db604acd2dded356603b0aa"}
{"id": "train_23337", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "ci_integration", "security_gates", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9a200c501fb823f68d1300f2ef1219f5fcfd4b9e"}
{"id": "train_23338", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "reproducibility", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fcf6ef9260dee6c58e9378b177472f60d9b49a85"}
{"id": "train_23339", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "tests_are_truth", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9c06a9d4235a1a7b26cc50c62a01651fe241f724"}
{"id": "train_23340", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "ci_integration", "auditability", "tests_are_truth"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c4d5de32fb29961b3b6e6193578b973dec524840"}
{"id": "train_23341", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "repo_scale_reasoning", "security_gates", "reproducibility"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "695ac3cfa5c8d7b2aaf6ad107183be0461fa3457"}
{"id": "train_23342", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "repo_scale_reasoning", "tooling", "security_gates"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "53d5c25a770e210f2aba897e9b15ffd78e9a9544"}
{"id": "train_23343", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "cost_latency_tradeoffs", "security_gates", "ci_integration"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1c7bd9042a1662e4ffc61d858d1598923823c7a4"}
{"id": "train_23344", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "governance", "documentation", "tests_are_truth"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c4dafbb0b8c5adaa25182f25e2ddd621a88dd80e"}
{"id": "train_23345", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "documentation", "tooling", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "967e61ad7271b82e249301d297fa28f1ed893eeb"}
{"id": "train_23346", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "documentation", "evaluation_metrics", "reproducibility"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a8a3b92f322f34bf1cbdc818e31b905d0a4687c7"}
{"id": "train_23347", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "governance", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a397bbd66c73b683f25e10c49e6af49741074ceb"}
{"id": "train_23348", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "reproducibility", "documentation", "tests_are_truth"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cb4a7506a52497c28c1ec9aad27b5e2115ca49cf"}
{"id": "train_23349", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "governance", "repo_scale_reasoning", "security_gates"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "df1b76f7dd496e511df59ed78251dca96688a3a7"}
{"id": "train_23350", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "repo_scale_reasoning", "tests_are_truth", "ci_integration"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f5978efff9c220e5cd69c49941851bb4afad5437"}
{"id": "train_23351", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "tooling", "security_gates", "ci_integration"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d0b29cd9274946aca8fe43ad59dc1eb91e853862"}
{"id": "train_23352", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "governance", "tooling"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b674a37da16fb7f860c6a083094785e985a302a9"}
{"id": "train_23353", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "security_gates", "auditability", "evaluation_metrics"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9e522dda307d848a8a567c9c0468a49f6665526f"}
{"id": "train_23354", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "tooling", "documentation", "auditability"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "153bbbe60a6ae1efb721cc469501cad658d801f4"}
{"id": "train_23355", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "security_gates", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "325de52976c5b2345d9456b94259b97044cdaf72"}
{"id": "train_23356", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["governance", "evaluation_metrics", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "58e03aff500ba570744e37ce02a761823ddfe220"}
{"id": "train_23357", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "tooling", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "52860bcd1918ed358937c481ca40674281c12d8c"}
{"id": "train_23358", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2684f44cf4ea522415d5342aadeabbec17f5db6f"}
{"id": "train_23359", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["governance", "security_gates", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "198dd560bda3e99b971716f20c5e4eaef7a406d0"}
{"id": "train_23360", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "governance", "tooling", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "170fd55efef58af499a35c6e494ea935aab59bdd"}
{"id": "train_23361", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "evaluation_metrics", "tests_are_truth", "security_gates"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c400ebdba4e511ca70eba95f2423840c8b52423e"}
{"id": "train_23362", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "tooling", "reproducibility", "ci_integration"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "015db741bcbfb6e4aab020966f5f5e585ca74774"}
{"id": "train_23363", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "tooling", "ci_integration", "reproducibility"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "56428b3d45c85227a1554df63d90c0e4b4712829"}
{"id": "train_23364", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "tooling", "security_gates", "ci_integration"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "62622fca2911a58f244aeaabe18f06439a2be1d3"}
{"id": "train_23365", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "reproducibility", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6851a9d1b24ed11ab2d03856594e04a7d24f7889"}
{"id": "train_23366", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "auditability", "reproducibility", "documentation"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "343394c7533375e2a57a6a4c8c1df5be42a675a7"}
{"id": "train_23367", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "tooling", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e9812d9bfedf8a8d7d429c750197d5c47aa5e53c"}
{"id": "train_23368", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "tooling", "reproducibility", "ci_integration"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "492ba6d8500e00f03fc3dd08024f24f7423abc9a"}
{"id": "train_23369", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "ci_integration", "repo_scale_reasoning", "tooling"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "56e6b5ee9851c35d53292c03fc8065e6f906a57d"}
{"id": "train_23370", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "repo_scale_reasoning", "reproducibility", "security_gates"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6075a0a602f4477f5ba5860be82b5a5e32578874"}
{"id": "train_23371", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "ci_integration", "security_gates", "reproducibility"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0ec3ea99161c933da64f3c4ee947e01542cf23ef"}
{"id": "train_23372", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "ci_integration", "evaluation_metrics", "documentation"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "863363bb0116dc036835424f8a9f51036b513fdb"}
{"id": "train_23373", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "tests_are_truth", "auditability", "tooling"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b3431b57e8695453fcd0ffea7bf6463c948ef17c"}
{"id": "train_23374", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "evaluation_metrics", "tooling", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "36ccb46ef56c2b946b910677b43cf890c54d6444"}
{"id": "train_23375", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "tooling", "governance"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b505dc572759e601517fd0f22ad7f0f09c0086d3"}
{"id": "train_23376", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "tests_are_truth", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2220597d8d90256f9d2505349a043bf7074342ba"}
{"id": "train_23377", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "cost_latency_tradeoffs", "auditability", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8ff94c75f34d6585bfc3fee6cd18bd881e7b894b"}
{"id": "train_23378", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "tooling", "governance", "evaluation_metrics"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a4026dba17233aa97a04fcd14431689e88345c4b"}
{"id": "train_23379", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "governance", "security_gates", "ci_integration"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d18b71f12497cefbfe9ca8546d541253f66a08c2"}
{"id": "train_23380", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "auditability", "reproducibility", "ci_integration"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "10fc2bff648a6353a07ba8e35e60c6cf5e46d1fc"}
{"id": "train_23381", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "reproducibility", "auditability"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "eaa1b8ec4e5fa44870560ca5bc76674bea0703f6"}
{"id": "train_23382", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "auditability", "repo_scale_reasoning", "security_gates"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "48e5ab8aec925b68d7c5c7ece8afab9d5f509ac0"}
{"id": "train_23383", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "tooling", "reproducibility", "governance"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "541bd0783867762e8dd2dacdc3059697424668ad"}
{"id": "train_23384", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "reproducibility", "repo_scale_reasoning", "tooling"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7037aa64f69b49bf12c9e1d3ff53de350606cde2"}
{"id": "train_23385", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "cost_latency_tradeoffs", "governance", "ci_integration"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "995f0dab2c7b6e09ed7994fbd1dcfe772e3d931d"}
{"id": "train_23386", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["auditability", "ci_integration", "tests_are_truth", "governance"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4c49abc24076dce99d757a3b6b7f4d8b1fcd766c"}
{"id": "train_23387", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3ce5463436cb8a3f61ef8d269a16e3b7d0d9993f"}
{"id": "train_23388", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "reproducibility", "evaluation_metrics", "ci_integration"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "010d23b9c5e4b4ecb9baa878b64a0d9903b5f34b"}
{"id": "train_23389", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "reproducibility", "governance", "ci_integration"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "228c75f4f83dded80b30a8488478ebfdc29f1e73"}
{"id": "train_23390", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "governance", "repo_scale_reasoning", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "74a386662ed22879cd00aec239650f8ea2aa8887"}
{"id": "train_23391", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "reproducibility", "tests_are_truth", "documentation"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "26a35abf1ae588e3cd91cf96dd8bbec2ffb5fae7"}
{"id": "train_23392", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "ci_integration", "security_gates", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "585ca8870d0b20d6603e23121f0b07163d5a5f7c"}
{"id": "train_23393", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "auditability", "repo_scale_reasoning", "tooling"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d6015ae33fa8444b36a75079fcdb73735ad8bc51"}
{"id": "train_23394", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "governance", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "88329af195baff8ce55aa245a2810f68649993c4"}
{"id": "train_23395", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "cost_latency_tradeoffs", "evaluation_metrics", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "da0dc4c8b4fca87099e6fe5e44380afe224afaa1"}
{"id": "train_23396", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "security_gates", "documentation", "reproducibility"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "90688174fffb71eac4c0285c4d03745087785cd8"}
{"id": "train_23397", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "evaluation_metrics", "security_gates", "documentation"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a7571c5f2978e2ddb81cc4e135f08942cca07cb6"}
{"id": "train_23398", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "ci_integration", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "79edc3176291c6725cb1ff91c342a0c23c8c6c05"}
{"id": "train_23399", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cd692d6294575f3a11aa46bf6adf503d1d3a1480"}
{"id": "train_23400", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "security_gates", "evaluation_metrics", "reproducibility"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4b7652584b3a3e572b43871f92203736c465cdc4"}
{"id": "train_23401", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["governance", "evaluation_metrics", "security_gates", "tests_are_truth"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "594e94c7cb3ead7f93d01bc504037e0d6aec6d82"}
{"id": "train_23402", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "auditability", "security_gates"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f1454f199bf24602ef7f301df360423c031b7ce0"}
{"id": "train_23403", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["governance", "tooling", "security_gates", "tests_are_truth"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4305e74bcc576ed5b241daf2fcf0bb97e5fb3849"}
{"id": "train_23404", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "ci_integration", "reproducibility"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e265113e0a53e4f981d00086becbb875e4b2a91e"}
{"id": "train_23405", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "repo_scale_reasoning", "cost_latency_tradeoffs", "security_gates"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6aadce92c9cb280d76b208bcba576f801a681126"}
{"id": "train_23406", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "security_gates", "documentation", "auditability"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f684539b29783c4ac6a1d5ea55f635ccd1abe91e"}
{"id": "train_23407", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "repo_scale_reasoning", "tests_are_truth", "security_gates"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9e951ac0833b2df5d32334ecb630b561dc094722"}
{"id": "train_23408", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "evaluation_metrics", "tooling", "ci_integration"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "299b4b81948b2db0d72b90f98c6d022a0e01240c"}
{"id": "train_23409", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "cost_latency_tradeoffs", "tests_are_truth", "reproducibility"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dec44cc5b7f03d6c332ef86bda30481b83ed56e0"}
{"id": "train_23410", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "documentation", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "728d24ba9f9fbc80401333d1446cd92fcc1fd414"}
{"id": "train_23411", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "tooling", "documentation", "reproducibility"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0fab73c9817fc33e1c2354da686ac5de2b74cbee"}
{"id": "train_23412", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "security_gates", "ci_integration", "governance"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "847250183ca689087e14a10de96fb7ac290a2af4"}
{"id": "train_23413", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "documentation", "evaluation_metrics", "auditability"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "73d53c9590ee4deeac43dc281a4d088c8effcf82"}
{"id": "train_23414", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "documentation", "reproducibility", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "27ade62b1a4083f062ce61f9254371c3cb96794c"}
{"id": "train_23415", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "evaluation_metrics", "security_gates", "ci_integration"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0da98d1a1f37b4647ed4d639915602a08f50fba7"}
{"id": "train_23416", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "auditability", "security_gates", "evaluation_metrics"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c11444599bb8a86cc409a791fe8aba950a94adf8"}
{"id": "train_23417", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "security_gates", "ci_integration", "tooling"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dcee2f409367b7cbe9a7211981e527d639b797fb"}
{"id": "train_23418", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "repo_scale_reasoning", "governance", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "51f0e6b16be5e50b4ea4314838a3e10713dc97ef"}
{"id": "train_23419", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "tests_are_truth", "reproducibility", "documentation"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7a44f6d21745e94ee6cf252362cbe7c75a4a0205"}
{"id": "train_23420", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "security_gates", "tests_are_truth", "auditability"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "834a78c4fca2f9552aaa89d7b475c0c028a499fc"}
{"id": "train_23421", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "evaluation_metrics", "governance", "ci_integration"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "29b021e0fdbf7b15864382b1729991b26484f94a"}
{"id": "train_23422", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "ci_integration", "evaluation_metrics", "governance"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7a203f7ff8176ffd4bfbaa3a0a1dc19cbc019b92"}
{"id": "train_23423", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "security_gates", "repo_scale_reasoning", "governance"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fc7f6ee73f5af258aca64fd245ef0a17a68cc421"}
{"id": "train_23424", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "governance", "security_gates", "ci_integration"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7dcd11194239811c92896469931eef0ce503c501"}
{"id": "train_23425", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "repo_scale_reasoning", "evaluation_metrics", "auditability"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c94b16ce36e1f579d4fff16241ec784e8dd552fd"}
{"id": "train_23426", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "tooling", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "36b356357d58e97e6feea84bf0bafd9d317d8791"}
{"id": "train_23427", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "repo_scale_reasoning", "ci_integration", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b742721083fad8e02cbbef4e5eb0b561afa09d15"}
{"id": "train_23428", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "cost_latency_tradeoffs", "governance", "evaluation_metrics"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d2fb2af9b00b0e74b6a37db50c69db42a34d1e0d"}
{"id": "train_23429", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "reproducibility", "repo_scale_reasoning", "evaluation_metrics"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "53d5c25a770e210f2aba897e9b15ffd78e9a9544"}
{"id": "train_23430", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "security_gates", "tooling", "tests_are_truth"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "87a5263c3f1344c6a407aabe57d92c15c43a8f03"}
{"id": "train_23431", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "auditability", "ci_integration"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "aa55f6ab9e6fcba518bc5eedec85b2d1ddae41a9"}
{"id": "train_23432", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "security_gates", "reproducibility", "tests_are_truth"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5f658bd54c134f24da9ded721e909b257f49da44"}
{"id": "train_23433", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "governance", "evaluation_metrics", "security_gates"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a7fb2b2c245c6106485f86af8c6aecf05a9d2016"}
{"id": "train_23434", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["auditability", "reproducibility", "evaluation_metrics", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "805b7a1c28cbeef13fca12dc3b55bd75bfa81749"}
{"id": "train_23435", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "ci_integration", "reproducibility", "auditability"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fc93a4cf5d56079ac98f3c8675fdaaf4bab536ec"}
{"id": "train_23436", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "tooling", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b095922fbf5166761c30e1ff9c54ff72fbd33e71"}
{"id": "train_23437", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "repo_scale_reasoning", "reproducibility", "documentation"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "17dbd1e5d118c0dda985909f156919afbfdc0f5d"}
{"id": "train_23438", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "cost_latency_tradeoffs", "ci_integration", "auditability"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "64cad735fbe618bf73e9290f6a77e521b004e0bb"}
{"id": "train_23439", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "reproducibility", "security_gates", "governance"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1ffbf524ec6e55953aa5d5fb57bfcd7afc53626b"}
{"id": "train_23440", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "ci_integration", "documentation", "reproducibility"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cd8b98b98104837ba45f274372b7806077b12c66"}
{"id": "train_23441", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "reproducibility", "security_gates", "auditability"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8a90d46d4dd158f7e100962302d6398368e222b6"}
{"id": "train_23442", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "governance", "tests_are_truth", "ci_integration"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "740610224de50d51e86589690140bcc4ec7f1758"}
{"id": "train_23443", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "governance", "repo_scale_reasoning", "auditability"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "735db2e5f6c2c3163491b5bc41dc3b7e0b91e178"}
{"id": "train_23444", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "governance", "reproducibility", "auditability"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "27574483c0e3f5f3c221a07971897e65bd207d47"}
{"id": "train_23445", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "evaluation_metrics", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b19a8861ef169749f51957eaf98f91ca00eecb52"}
{"id": "train_23446", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "tooling", "repo_scale_reasoning", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f7987a1ce7029ef66e90ec994772b0219c780eee"}
{"id": "train_23447", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "auditability", "documentation", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a58bb3cabebe3cb5ceca156ce2b224930fc21d27"}
{"id": "train_23448", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "tooling", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "15286a6f1970ad477d2065e59c1c3c31f09c5826"}
{"id": "train_23449", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "evaluation_metrics", "auditability", "repo_scale_reasoning"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fac0c71ad881aa47f07b74e1984d340ec4d88f50"}
{"id": "train_23450", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "repo_scale_reasoning", "documentation", "governance"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5f30b6ba9ce9f8f200fc1db340395d69a5234fd6"}
{"id": "train_23451", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "ci_integration", "tooling"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d516bfa2c1db9fb66cdc193dc9ccbf34d3c25825"}
{"id": "train_23452", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "documentation", "reproducibility"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ede7fc5bb114ffb3a510e76ae03dae122812d496"}
{"id": "train_23453", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "ci_integration", "auditability", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dc735220db6fe5787d570ddd1d27136572876bf9"}
{"id": "train_23454", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["governance", "security_gates", "repo_scale_reasoning", "documentation"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9651cc6493c4fdeeb71f51c20a552b9d49787b80"}
{"id": "train_23455", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "documentation", "governance", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d0c69f56440eb20237ccf481ad4c289d43edb734"}
{"id": "train_23456", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "security_gates", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b7deeafb6097c3bb3532dafaa9336412e23e1705"}
{"id": "train_23457", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "cost_latency_tradeoffs", "ci_integration", "security_gates"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "208d866ce708b2863a4f574d8519cbd77fa38b22"}
{"id": "train_23458", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "repo_scale_reasoning", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7de2f80ded34c23528a585f09311dfb040a5075f"}
{"id": "train_23459", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "ci_integration", "documentation"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "609c53633ed2cc990a72fe90e706769e31bd15f5"}
{"id": "train_23460", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "tooling", "evaluation_metrics", "reproducibility"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6c4177cf3a63c6cf516d66244f83011c977c0811"}
{"id": "train_23461", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "auditability", "security_gates", "tooling"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bd1fc3b2405e8204299b1a368af34332d39fa716"}
{"id": "train_23462", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "auditability", "governance", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d01165a3fa8cae06f1ba3a50a54c9323dd027042"}
{"id": "train_23463", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "evaluation_metrics", "governance", "repo_scale_reasoning"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3c510acc8647790680160e02d240281c52a95935"}
{"id": "train_23464", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "reproducibility", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8c88f96d4a4f4724ef3a1a5fd285058e255b5057"}
{"id": "train_23465", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "evaluation_metrics", "cost_latency_tradeoffs", "security_gates"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "de28226c0843cead6a26f2399f56eeb5c5a89d4a"}
{"id": "train_23466", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "evaluation_metrics", "governance"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "18a185fecd9a475d5bcc2258ae94f03fa8a52cd8"}
{"id": "train_23467", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "evaluation_metrics", "reproducibility"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fc09ba550aa38207709c07eb95f3f227ec875d5c"}
{"id": "train_23468", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "documentation", "auditability", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "90b2a16e93076f77bbb8db7c5ae30bc80e595c5f"}
{"id": "train_23469", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "ci_integration", "governance", "tests_are_truth"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "54d18503d613555fff89a56ced5526a028eb7207"}
{"id": "train_23470", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["governance", "cost_latency_tradeoffs", "security_gates", "ci_integration"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0cd681c58f597c92d603d0db721054a75025ae36"}
{"id": "train_23471", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "reproducibility", "tooling", "documentation"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f3119d0afdd4002b6ad8bbedb34605dbcda64cff"}
{"id": "train_23472", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "ci_integration", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "803a7afb533ac2369a29a9b87b843976b874b3a2"}
{"id": "train_23473", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "governance", "evaluation_metrics", "security_gates"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cf8a50e9b7571398c77f785e707ef1b71f834dbe"}
{"id": "train_23474", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "tooling", "ci_integration", "tests_are_truth"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "efe1e0a9189eff8843e65a384cc6b17c5ddbbe25"}
{"id": "train_23475", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "governance", "cost_latency_tradeoffs", "security_gates"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bb19004c47e5027b7bd2b6e3573a6df5108f664f"}
{"id": "train_23476", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "ci_integration", "security_gates", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f1df77f5a391707045d647b3b3350ae06abb20b1"}
{"id": "train_23477", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "governance", "ci_integration"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8c20182062b9bc0fd5eb5472c5cfd07372eac55f"}
{"id": "train_23478", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "reproducibility", "tooling", "security_gates"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "42057a934e945af10faf0d22d83376241455e967"}
{"id": "train_23479", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "governance", "reproducibility"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "feab74f6d5acf8a01640e0714ddb649843e3824a"}
{"id": "train_23480", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "governance", "ci_integration", "tests_are_truth"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "abe46f0535c3e40f50867ed7948c127be7914e4e"}
{"id": "train_23481", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "documentation", "auditability", "ci_integration"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7db75282fc9df12cd5db6a75e0ad9eef7310b73d"}
{"id": "train_23482", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "evaluation_metrics", "governance", "ci_integration"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5d745f842aaff0cb53bb09e8335cd46e025ae338"}
{"id": "train_23483", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "repo_scale_reasoning", "governance", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8acc777972f3e61189ede900d63126b49e92c98a"}
{"id": "train_23484", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "documentation", "ci_integration"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bfd5446ad75b418567355c28e5ba5cbf52dc7006"}
{"id": "train_23485", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "repo_scale_reasoning", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "972e45e41c3a1b3886f608b967696026a26c76f8"}
{"id": "train_23486", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "evaluation_metrics", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "422a4ae4edfe26be06ca5fe4b45b286eeb893a0b"}
{"id": "train_23487", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "security_gates", "auditability", "evaluation_metrics"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "91156cfe77e02c17444a9e1ec9c082d60c0a6f7f"}
{"id": "train_23488", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "evaluation_metrics", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4ea35580e409eb99bb5a99e17254c1ea1ed9eb32"}
{"id": "train_23489", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "reproducibility", "tooling"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1e85b8008fb78fa3ba0bd6a5431bcc5b6027c08a"}
{"id": "train_23490", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "repo_scale_reasoning", "auditability", "tooling"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3d0ee69119a9a505fc4231c2bed742f609e11c39"}
{"id": "train_23491", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "repo_scale_reasoning", "documentation"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "13381b03849540884ab727b6d34f4ef97d541550"}
{"id": "train_23492", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "repo_scale_reasoning", "auditability", "governance"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c445379f6ef961a7b41fbb2685806a914d6f14a5"}
{"id": "train_23493", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "auditability", "reproducibility", "security_gates"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "07827913092903ae38050fd5f47334da479d34ab"}
{"id": "train_23494", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "documentation", "security_gates"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "504f6a57807f48ae9cdf0c84223f795844ed27f1"}
{"id": "train_23495", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "tests_are_truth", "tooling", "security_gates"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "05c6cb390ef6183649ff613ee23a0979918ec671"}
{"id": "train_23496", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["documentation", "auditability", "tooling", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "41b92495c66e5021d19fa6fb15f9a226017bec42"}
{"id": "train_23497", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "ci_integration", "evaluation_metrics"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d36a2ebbeaaaed0adc828cbe56fc9dd41c845b13"}
{"id": "train_23498", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "documentation", "repo_scale_reasoning", "tooling"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "36ab26cc72c496e1cff34b9ff344034d6fbb01d6"}
{"id": "train_23499", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "governance", "evaluation_metrics", "documentation"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "07ea788746c8c62781b75c2a0d6ea352071ca69f"}
{"id": "train_23500", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["documentation", "tests_are_truth", "tooling", "ci_integration"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8e56e6e13449e15ab8e8a0a8e45b1a6e867c3987"}
{"id": "train_23501", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "repo_scale_reasoning", "tooling", "ci_integration"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e9f3f4c05dfa065f3336893f30a2e205b1ea9eaa"}
{"id": "train_23502", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "tests_are_truth", "repo_scale_reasoning", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "08683bcffa9c3403019a09918c60dfc920577b45"}
{"id": "train_23503", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "tests_are_truth", "repo_scale_reasoning", "auditability"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3b3b4832075e79f629c66a9970dc052db4380a17"}
{"id": "train_23504", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "auditability", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "eb38647dd10b09768a8c05c6c978072ccc56dc44"}
{"id": "train_23505", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "reproducibility", "tooling", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8ce51f9cae436ca38998375dc96cc47fffc5cf16"}
{"id": "train_23506", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "security_gates", "tooling", "governance"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fbb2f13a0aa8b2e5303e0b3a8f2133d7a91178e5"}
{"id": "train_23507", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "security_gates", "tests_are_truth", "governance"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b8ba36120577e346a20239cfad8fdd4cddd08937"}
{"id": "train_23508", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "auditability", "tooling"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5c8a21dbcae050e21fa43027805685aa21889513"}
{"id": "train_23509", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "security_gates", "ci_integration", "tooling"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "627078819912028cec810f60e01bb27ab29be08a"}
{"id": "train_23510", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "tests_are_truth", "cost_latency_tradeoffs", "repo_scale_reasoning"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5fd95876479ac939780ad6367b6403d87105e3e6"}
{"id": "train_23511", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "ci_integration", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5d67d9300656f36db54432fe4935d8cf8304b454"}
{"id": "train_23512", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "governance", "documentation", "security_gates"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "227154eab4121793b56be8b353688e615909de5e"}
{"id": "train_23513", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "cost_latency_tradeoffs", "auditability", "reproducibility"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d881b93b9c6bdf00bb58eb39918ecc368f4c8b95"}
{"id": "train_23514", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "reproducibility", "governance", "evaluation_metrics"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "98b70db750ce2dd0b1a957016160ecb6fb169b5b"}
{"id": "train_23515", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "auditability", "tooling", "reproducibility"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8573db12577bc47d449d78a0044464068155ca5b"}
{"id": "train_23516", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "ci_integration", "auditability", "security_gates"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c4d70c3494b2cf0c929137834d23a699420d2219"}
{"id": "train_23517", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "tests_are_truth", "documentation", "security_gates"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "701bfcf700bd40e0a825f5d77feb151b5dd7b6ba"}
{"id": "train_23518", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "ci_integration", "tooling"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e0b1cafddcad3ef915d2dd99265307c529f97158"}
{"id": "train_23519", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "evaluation_metrics", "documentation", "tooling"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a7e8d375823eb96e683841a0a54ac2d8d4e3beeb"}
{"id": "train_23520", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "reproducibility", "auditability", "tests_are_truth"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ff18962bad8fe65191ff769ebf58dae81fa15e35"}
{"id": "train_23521", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "ci_integration", "documentation", "reproducibility"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f086ca00037f29734852e94c150c7cd20ac723f0"}
{"id": "train_23522", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "reproducibility", "ci_integration", "tooling"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5be1ab91785502645fff2276da2106983721d24f"}
{"id": "train_23523", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "ci_integration", "documentation"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cea301a61d018ca6d74dda16b470000cca3bbc3d"}
{"id": "train_23524", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "tests_are_truth", "ci_integration"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c95d4894a695bd6c7767d1495417a00e36997a7e"}
{"id": "train_23525", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "reproducibility", "tooling", "documentation"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "46474ce29984e790e93c8e55495d95315cfae218"}
{"id": "train_23526", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "ci_integration", "tooling", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "673dc9fd76b198dda9cb38a63037a4799f1aa89e"}
{"id": "train_23527", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "security_gates", "governance", "ci_integration"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "011ca5431e7a6c5f8b7801361640b6000db0a035"}
{"id": "train_23528", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "documentation", "security_gates", "ci_integration"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "16e4970ac029f15f85f8cd33e7c786eb4539e592"}
{"id": "train_23529", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "security_gates", "tests_are_truth", "documentation"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "01c681b1a5de98d750e2e4c82232df6113595ef2"}
{"id": "train_23530", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "cost_latency_tradeoffs", "tooling", "ci_integration"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "00f9416876f3c0d57b60891c557c1d7ad370e958"}
{"id": "train_23531", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "auditability", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "568f38e6ce6055300324b1d04db8e289fa9d3b2a"}
{"id": "train_23532", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "ci_integration", "repo_scale_reasoning", "evaluation_metrics"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "26d8627909420055e975fd8dc04ec0200cb2f85f"}
{"id": "train_23533", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "tooling", "evaluation_metrics", "governance"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "57c75dfdc0835f5663ce81fb7c247c8c1be5bcf5"}
{"id": "train_23534", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "tooling", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4cbb10c0793df64c1439eb44574bde2021e11a0d"}
{"id": "train_23535", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "ci_integration", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5a1a084ef74bfcb82781072c49105b13500f8ae6"}
{"id": "train_23536", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "documentation", "governance", "tooling"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fc63b2100539f752e6ba4f00cfed7bb3630d1a26"}
{"id": "train_23537", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "ci_integration", "repo_scale_reasoning", "auditability"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ca9fd6d9f890869349b3ac5892a8dc32357aacb4"}
{"id": "train_23538", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "tooling", "security_gates", "tests_are_truth"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e3212f7a2f987dbfd9968a44af416a599b02fd51"}
{"id": "train_23539", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "tooling", "auditability", "ci_integration"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "eeacd23311070bcc123aaf507b99d6e6bb6ea68e"}
{"id": "train_23540", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "reproducibility", "governance"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2093c0b9d150f01ef11a898da3f13923f9105e1f"}
{"id": "train_23541", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "tooling", "tests_are_truth", "documentation"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "010eeee1b4195e0f9f09e072f90d08280db83a1f"}
{"id": "train_23542", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "ci_integration", "governance", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bcad9d5f7885ecf03961aad75ec0e9a84301bac8"}
{"id": "train_23543", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "reproducibility", "documentation", "security_gates"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0f0b0ace0b7e4be9014f98c1645d5c7482194589"}
{"id": "train_23544", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "auditability", "tooling", "documentation"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b2dd3bf6b9f25ef9ac86ea13315390907ebf0809"}
{"id": "train_23545", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "reproducibility", "tooling"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "644f43bfa5f169b775a6d7e565a63bbaeaec811e"}
{"id": "train_23546", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "security_gates", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "69a09690fcb9f006a5938eadb051cc9444d7c6ff"}
{"id": "train_23547", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "cost_latency_tradeoffs", "tooling", "tests_are_truth"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "31e89c5e3a590b00753cc42741d2ff24a31394df"}
{"id": "train_23548", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "tooling", "evaluation_metrics", "security_gates"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "685c27f81bf4670125e4bb87d5c92484b0ebb24f"}
{"id": "train_23549", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "tooling", "documentation"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bcabf1caf7f9bbb836f5dfa6166b23e061bfa2c1"}
{"id": "train_23550", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "auditability", "documentation"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "52a208b209d08990f6bcb06b4b485df60b9e9e8d"}
{"id": "train_23551", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "cost_latency_tradeoffs", "documentation", "ci_integration"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ee8d13a33ad30d06bf046140346790495a7b1dbb"}
{"id": "train_23552", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "documentation", "auditability"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "55859329f2cdb58308986d553248737eb83936cb"}
{"id": "train_23553", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "reproducibility", "governance", "documentation"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5a7f131436910755f9e255f836f0af43ec919d83"}
{"id": "train_23554", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "repo_scale_reasoning", "auditability", "tests_are_truth"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c28c79520b0b5a9a350e432c0103ebae7a40c44b"}
{"id": "train_23555", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "reproducibility", "governance", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e40f11375e770c3da40981bcc8c03cf9c8f2089a"}
{"id": "train_23556", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "security_gates", "reproducibility", "tests_are_truth"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5aa1ab3e43f3cdf2c0eb5766adcd306e95ddf20f"}
{"id": "train_23557", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "auditability", "documentation"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6e61c11975c401200d67781de560835e9827bda2"}
{"id": "train_23558", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "tooling", "governance", "ci_integration"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3e230847e1356ed26fc97c6607de4fcf648703e9"}
{"id": "train_23559", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "governance", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d75ddb794870734285aff369595ddaecabccb688"}
{"id": "train_23560", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["governance", "auditability", "tooling", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d38904ef44ba56d6c494c94f450c17bbef4de532"}
{"id": "train_23561", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "documentation", "security_gates", "ci_integration"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9dadb9fd376d0d3f935b2e83e5b01ac82cdab0a6"}
{"id": "train_23562", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "governance", "evaluation_metrics", "ci_integration"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "561006c1208e2ad90f728174779944d1cf762bec"}
{"id": "train_23563", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "auditability", "governance", "ci_integration"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9601aa47b60f711a52b4f945663adb1eba103c26"}
{"id": "train_23564", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "repo_scale_reasoning", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c477bb993d55082cb8605ae3ae2d4dcb5f2ae692"}
{"id": "train_23565", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "tests_are_truth", "tooling", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "992fb59989bfb5401162e45b5ff327c521a590ea"}
{"id": "train_23566", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["governance", "ci_integration", "auditability", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "22367942ebd91115ff1df5cda1a3c1af6f57e7f1"}
{"id": "train_23567", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "tooling", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5a7479e114f9a5f8e57f60ad9eafb6b64f0f6b4b"}
{"id": "train_23568", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "ci_integration", "documentation", "reproducibility"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "088f62e5df002ddc01970f7b396c196812bb85c9"}
{"id": "train_23569", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "security_gates", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ddda3a31cac6ce880112ac4b8b8c04a3ee5f9ff2"}
{"id": "train_23570", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "repo_scale_reasoning", "tooling", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b32329819b70682398c07256e50c097ce56bb2b3"}
{"id": "train_23571", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "evaluation_metrics", "reproducibility", "tooling"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "42e9b3ee74507e5fb3db7a71888bbb03b6bc0cc4"}
{"id": "train_23572", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "reproducibility", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c6dd67450c48ee6179a3d02ec3b2d1a6964b2862"}
{"id": "train_23573", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "repo_scale_reasoning", "reproducibility", "tests_are_truth"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "37926d4ed4915e942f589f4097ba7fda892274fb"}
{"id": "train_23574", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "governance", "evaluation_metrics", "documentation"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f27d9bdfa6b25703a863f21a90ba21ec9f24fdcd"}
{"id": "train_23575", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "repo_scale_reasoning", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6c07acf296451868c58d692b1d6157a770314e35"}
{"id": "train_23576", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "documentation", "security_gates", "auditability"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9c03d1c621c1a8c86698a399d61178a48ca4a6c4"}
{"id": "train_23577", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "governance", "documentation", "reproducibility"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "10b4d81aa8639602c71db2855dd9ebb8d6ce0a92"}
{"id": "train_23578", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "security_gates", "auditability", "documentation"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5461cb49d037e4154c2924b1ed812e2c5ad5085c"}
{"id": "train_23579", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "documentation", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d7ca010f0dcf04c39d0b6eb8be18d3bb1cf5260e"}
{"id": "train_23580", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "repo_scale_reasoning", "governance", "reproducibility"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "08906fc1af6bfe02b620e1051db49bb0701062cf"}
{"id": "train_23581", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "auditability", "evaluation_metrics", "governance"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a9fb2840dd51585443265d32faf15ebe3282e6e5"}
{"id": "train_23582", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "documentation", "repo_scale_reasoning", "governance"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "66fb9fd1c041d6f3cc229b738aa02309f5e0c2a3"}
{"id": "train_23583", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "auditability", "tooling", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7543a8ae8973a57de726d90ae36f763a5797de8c"}
{"id": "train_23584", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "reproducibility", "security_gates", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "917d81a3a28624f0f37c18571c83911fef5db450"}
{"id": "train_23585", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "security_gates", "evaluation_metrics", "tooling"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4317206211ec00d236e7a5b81c12393a59206c89"}
{"id": "train_23586", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "tests_are_truth", "ci_integration", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b0bc1eec282f904a404c17dd53360b062aa7c364"}
{"id": "train_23587", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "ci_integration", "evaluation_metrics", "reproducibility"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "29cca41f840ab96da59979d581125b8fb82dc8de"}
{"id": "train_23588", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "tooling", "security_gates", "tests_are_truth"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a1a63708361d58bae0588f0979a5758eeb639200"}
{"id": "train_23589", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "governance", "auditability", "documentation"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b36320586a594b503cb598666af1d53ed73ef3f0"}
{"id": "train_23590", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "documentation", "ci_integration", "tooling"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ca230a0fc821b9d2f3bc08eddb71ba92926d05e6"}
{"id": "train_23591", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "tests_are_truth", "auditability", "evaluation_metrics"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "81b97909062fbf3c9e068032042069a5d47c2d1b"}
{"id": "train_23592", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "auditability", "ci_integration", "documentation"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c7c15bbda525a7fcbadb92270d7735aaceb45e09"}
{"id": "train_23593", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "tests_are_truth", "documentation", "auditability"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "66a5dad5e5c759f4beaaec3770e749e643d22370"}
{"id": "train_23594", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "ci_integration", "documentation", "governance"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1b0152fc867cb73ce7f0a370c4069462bcf7cbc2"}
{"id": "train_23595", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "evaluation_metrics", "tooling", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d7fe4f2db975bafad3673d44020efe411eb51c55"}
{"id": "train_23596", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "evaluation_metrics", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8fe15aae8c00e7055050a921e9b07acdb59c2520"}
{"id": "train_23597", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "governance", "evaluation_metrics", "security_gates"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5cf4e04e5f18fd953e69d380ddf8a75c140fa0b7"}
{"id": "train_23598", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "governance", "ci_integration"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e69440fa48c6c7445997b94f18b591d88407a3ff"}
{"id": "train_23599", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "ci_integration", "evaluation_metrics", "tooling"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ef83aae2009311eaadf0e4e5342ccdf20c30c6ff"}
{"id": "train_23600", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "tests_are_truth", "ci_integration", "auditability"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "54a87ac9719b6cc4a17a6961aecacf6b05b28cc7"}
{"id": "train_23601", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "security_gates", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f16bc08663bff10cf2d963950ad5b78f45371901"}
{"id": "train_23602", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "tooling", "ci_integration", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "41978654eeee9fe1683f532a64f0ed31e3ed2d6c"}
{"id": "train_23603", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "governance", "documentation", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "022cf2df28d9c9554fbc05ef7b8228d75d943fdb"}
{"id": "train_23604", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "evaluation_metrics", "security_gates"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1d386b05144bcd04c95ef51ac362f1b13138754d"}
{"id": "train_23605", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "repo_scale_reasoning", "tooling"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "899e6675c618b51c39e8841edaca0caa4b764a5a"}
{"id": "train_23606", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "governance", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cf9c6b4cf431b7e69f4031fe1def6d5306819807"}
{"id": "train_23607", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "auditability", "ci_integration", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4438a7c467c1c765393bb36c59d6b193f6d2fff4"}
{"id": "train_23608", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "repo_scale_reasoning", "security_gates", "governance"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a2aaed0f0d579c4d0d966d7ea21578581fc4e9c2"}
{"id": "train_23609", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "tooling", "repo_scale_reasoning", "evaluation_metrics"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bd6300f4313b4713c4e975a56133b70ef13e1a40"}
{"id": "train_23610", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "evaluation_metrics", "auditability", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d59addf2c18f6619ea8c150d07e4a8d59a9de049"}
{"id": "train_23611", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "governance", "repo_scale_reasoning", "security_gates"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "70253284ea296b08a7b1f90df660e7c8531a462c"}
{"id": "train_23612", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "reproducibility", "tests_are_truth", "auditability"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e92d8d249386f2c8313cf3bef20a6df1316cbe85"}
{"id": "train_23613", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "auditability", "documentation", "security_gates"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "221d8d5b52a79906b89fed2ee0563fba439ddc29"}
{"id": "train_23614", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "tooling", "auditability"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "98903f42fbda2bbb060901d1476e0bc52d9414aa"}
{"id": "train_23615", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "ci_integration", "governance", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "22d8023af9ff7ad1656ec76a6132525aaa286534"}
{"id": "train_23616", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "reproducibility", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c0dfc7525da3d12fbefa36df45338e1c56105539"}
{"id": "train_23617", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "ci_integration", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a6a501cde4451f9e4f8de0f9a478299776547336"}
{"id": "train_23618", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "reproducibility", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c313040843d4f031b5adcfd3ebdb606118d8cf6e"}
{"id": "train_23619", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "security_gates", "reproducibility"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "aa1db1df4d64f97f3ae6709757f75d686de161b4"}
{"id": "train_23620", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "security_gates", "ci_integration", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b3f61bdfd17e240a0a7e69ffd9334dc1037c273a"}
{"id": "train_23621", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "governance", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "79d29f47642c47192af5cda7adae65bdb8036933"}
{"id": "train_23622", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "tests_are_truth", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5595c1cb220ee5c6c2cfde50a20df11af6ff7726"}
{"id": "train_23623", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "auditability", "evaluation_metrics", "reproducibility"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2052b0f1f3225f1bd669e55ab96599f868c2aca1"}
{"id": "train_23624", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "governance", "documentation", "auditability"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "78ae79861b3d25e6f2aa754857b969280151dd8b"}
{"id": "train_23625", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "reproducibility", "tests_are_truth", "governance"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "da6386bd3d7224d748de85a748ab14675a2d063e"}
{"id": "train_23626", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "repo_scale_reasoning", "security_gates", "governance"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5bd73341c693fd9fb6b0945c86544dfc02324a1e"}
{"id": "train_23627", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "ci_integration", "security_gates", "documentation"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b259e04d0b1638648477cea1eff2abfc02cff300"}
{"id": "train_23628", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "auditability", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e032b2d3de634e79492d713cfb0f11c840b64984"}
{"id": "train_23629", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "reproducibility", "evaluation_metrics", "documentation"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9bf3097ef769219d511f1b3be31901d69f06b537"}
{"id": "train_23630", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "governance", "tooling", "evaluation_metrics"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7a83052652553942dc8cd854249fa8c3bd6a0ca3"}
{"id": "train_23631", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "governance", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fc2485ae7eaf55b2bcef59c724b90ba059085711"}
{"id": "train_23632", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "evaluation_metrics", "tests_are_truth", "auditability"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3682cb717e97be48e7853a32f4d93920e62015aa"}
{"id": "train_23633", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "tooling", "evaluation_metrics", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e33c8343d7911bf11baea281cbe1939839eb9a3f"}
{"id": "train_23634", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["auditability", "ci_integration", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "870940ef96d3a3d726323dd757ee08d744a20acd"}
{"id": "train_23635", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "reproducibility", "tooling", "auditability"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b5f3e85b9adb9698404bf6f19a027a153f522df8"}
{"id": "train_23636", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "security_gates", "auditability"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4d6f669eb9ae148417aa7f452c7d0c3d0878ed21"}
{"id": "train_23637", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "auditability", "security_gates"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "77dd87bdeaf98175e4e580a815adf86733ed4eae"}
{"id": "train_23638", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "documentation", "cost_latency_tradeoffs", "security_gates"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b2dd3bf6b9f25ef9ac86ea13315390907ebf0809"}
{"id": "train_23639", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "evaluation_metrics", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0e2b06065babe25734cadf81ca86e7ad6c6e3b8b"}
{"id": "train_23640", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "reproducibility", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bbb19da14ceca463fa83395f7d3a5679baa5e8d2"}
{"id": "train_23641", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "governance", "security_gates", "tooling"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9ae256fd5f56ff85a25fdb113dee6ecec015117a"}
{"id": "train_23642", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "tests_are_truth", "security_gates", "tooling"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5d5e95eafe4cdfcdc369a052aeefd2483dcfb8c8"}
{"id": "train_23643", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "documentation", "security_gates", "tooling"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6247c587df795c5c2c4fbdd0029a13c45c67f105"}
{"id": "train_23644", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "auditability", "ci_integration", "evaluation_metrics"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "319f89746c0ae0be9f83e39aba7ac3664bb3444f"}
{"id": "train_23645", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "auditability", "evaluation_metrics", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9359948ac38e70d017b991596f0dbf34ca9e9269"}
{"id": "train_23646", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "governance", "auditability"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c996019e77cb745c67be7bac2c4ee00f805cfef3"}
{"id": "train_23647", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "tooling", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d0e6efbb215a773c0b70762537e185a9f645dcc0"}
{"id": "train_23648", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "governance", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4be1cc58572185c9840d7e678bdfb4ae1e81e7b4"}
{"id": "train_23649", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["governance", "cost_latency_tradeoffs", "evaluation_metrics", "reproducibility"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5802a0f98c087a0c20648ae2ed809b286c4ee5f1"}
{"id": "train_23650", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["auditability", "evaluation_metrics", "tooling", "documentation"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0d75298a08b74428290946d61872fdcb71c265cc"}
{"id": "train_23651", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "evaluation_metrics", "governance", "ci_integration"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2220597d8d90256f9d2505349a043bf7074342ba"}
{"id": "train_23652", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "auditability", "evaluation_metrics"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "51502071c280db9df313f35300c23d58f358c102"}
{"id": "train_23653", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "reproducibility", "security_gates", "governance"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "896e653d4ad3c87eb0568655a8170f6f1f642392"}
{"id": "train_23654", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "tests_are_truth", "ci_integration"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5670d5802c646dba0f89d08d312fc369df7d67ce"}
{"id": "train_23655", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "security_gates", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4b2c6717d4d346141265daee7aad125371e636ca"}
{"id": "train_23656", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "reproducibility", "security_gates", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ddd02f85e5bcd42bf2e3dea1fcf6e4a9c60cf44d"}
{"id": "train_23657", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "reproducibility", "tests_are_truth", "auditability"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0164646ede2931d81de8e1166bbaa2670910158e"}
{"id": "train_23658", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "governance", "ci_integration"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "93f562ed0b739f146dd447b35f13b47be9519366"}
{"id": "train_23659", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "auditability", "governance", "tests_are_truth"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f5d8de4cc8fae721021f92121c6dedad34036711"}
{"id": "train_23660", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "tooling", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "917d81a3a28624f0f37c18571c83911fef5db450"}
{"id": "train_23661", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "reproducibility", "tests_are_truth", "tooling"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0d1b0f464d8d520dcaabefc6a4231e8f732a1e3d"}
{"id": "train_23662", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "tooling", "repo_scale_reasoning", "evaluation_metrics"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2af1e09adeb5f05da47e8897b0ff846af5f31094"}
{"id": "train_23663", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "documentation", "governance", "tooling"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "016f2e27d541a4b56ed5171fe9fbbc211bd42e98"}
{"id": "train_23664", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "documentation", "ci_integration", "governance"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "51599702f7b85ded1c2dae6f4d04626928ce96de"}
{"id": "train_23665", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "tests_are_truth", "evaluation_metrics", "reproducibility"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7b74d3ae7ef7080381dd85cbbeaf13eb1ea1de3e"}
{"id": "train_23666", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "governance", "documentation", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "16161d72e2f69a8f60e9ceb0ccda61748b77a31c"}
{"id": "train_23667", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "ci_integration", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "89a18bd6c93ee44490204485f3018d1046f3376c"}
{"id": "train_23668", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "tests_are_truth", "reproducibility"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fd67efb28b1a331d55f066c40e6c36b370d818e0"}
{"id": "train_23669", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "repo_scale_reasoning", "ci_integration", "reproducibility"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9f228a4c3de5c53993c4751581c0eafdfe0f25c3"}
{"id": "train_23670", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "evaluation_metrics", "security_gates"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "16b552cb3c069933ba578d0c64c22abd52305d6e"}
{"id": "train_23671", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["governance", "reproducibility", "repo_scale_reasoning", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "be18e824328be64f6d94f96e9b21839d1f4e9cd3"}
{"id": "train_23672", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "documentation", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "47b2628d34b4812f447101602173eb7aaafb5d97"}
{"id": "train_23673", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "ci_integration", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7de2f80ded34c23528a585f09311dfb040a5075f"}
{"id": "train_23674", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "reproducibility", "governance", "security_gates"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0342b4f910f682bab0371603080c7444be74334a"}
{"id": "train_23675", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "evaluation_metrics", "reproducibility", "auditability"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "742dc05b5f96b3491fa47c7cb4d7c4a3fa4f01ba"}
{"id": "train_23676", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "tooling", "auditability"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "58b7954a42fac6a2deaec3a2f577ae82c85d3c6a"}
{"id": "train_23677", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "reproducibility", "tooling", "documentation"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a38710fccda7ff23b7c68c0783ee0b6b39aef5ee"}
{"id": "train_23678", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["auditability", "reproducibility", "tooling", "security_gates"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e0cbc00ea784253fa2892fbfa98674092f6226ff"}
{"id": "train_23679", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "repo_scale_reasoning", "tooling"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cb8fbe885dbc69761128c9bb5d088ba5a5d67647"}
{"id": "train_23680", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "governance", "reproducibility"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6a81351edb899ed6fd02359f2d344f2b57fdffbf"}
{"id": "train_23681", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "evaluation_metrics", "repo_scale_reasoning", "governance"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "56c5131e04e3aca002cde0954525dc2d6fe26aa6"}
{"id": "train_23682", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "tests_are_truth", "cost_latency_tradeoffs", "governance"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cfe7aff556eadb3f5a3091f6fc4dbd78e210528c"}
{"id": "train_23683", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "reproducibility", "auditability", "governance"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9500c717c13236f2837f2cadaf17cbd26b01359d"}
{"id": "train_23684", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "ci_integration", "tests_are_truth", "documentation"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "11bbccf6e441bf5055ad82241a62ac46ffed24b9"}
{"id": "train_23685", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "governance", "repo_scale_reasoning", "documentation"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5f5ab17ef15c10340e8946a4ae297c82aba52813"}
{"id": "train_23686", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "auditability", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "be5c0302d3dcf89d75bdf6b6d8909b40c7dea8f9"}
{"id": "train_23687", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "documentation", "auditability", "security_gates"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7cbda16d4faa66659e0b6161f88a6eb9cf7895eb"}
{"id": "train_23688", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "security_gates", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7968ebfe77f2daa5250fe5c7816c4d1dbbeeb552"}
{"id": "train_23689", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "tests_are_truth", "evaluation_metrics", "documentation"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e58a234b9eda94cc4f68edbf0224e8f1339504d9"}
{"id": "train_23690", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "governance", "cost_latency_tradeoffs", "security_gates"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3d74931321744b25fecff8b48cb56f4857b3620a"}
{"id": "train_23691", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "evaluation_metrics", "documentation", "ci_integration"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "438e41be4677cd716440875351af88a91b5c36be"}
{"id": "train_23692", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "ci_integration", "documentation", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9caa372e0a3d7e4e2dc1d7ce55fecc3343de8d32"}
{"id": "train_23693", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "ci_integration", "reproducibility", "security_gates"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b6d8a217f2fda8bb0a9cb1b3b1a134a9f03fba31"}
{"id": "train_23694", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "governance", "ci_integration", "auditability"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f939bb0f7c4cc66c7a2cd4faab8bf59f4960d9b9"}
{"id": "train_23695", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "tooling", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "92a37f59bbaa9cffee440d96c6712ce44a2768e0"}
{"id": "train_23696", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "governance", "security_gates", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d01ce6445fcfaf5e6dbb4a99ac743f6f36621f67"}
{"id": "train_23697", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "auditability", "governance", "reproducibility"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8e411caaf4983aeb56b67262a165452f93bc3257"}
{"id": "train_23698", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "evaluation_metrics", "reproducibility", "security_gates"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2c65b8ee78d53c7a2936cb21b4f61b4df17951ea"}
{"id": "train_23699", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "evaluation_metrics", "auditability"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "817744671e204d1bcb4f5f2cc1108c42187777fa"}
{"id": "train_23700", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "governance", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "787e870cae58da2bca0cb25c6be09c8ac506be81"}
{"id": "train_23701", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "evaluation_metrics", "tooling", "security_gates"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8534ae78ecf4e49327adec448f8952a924586265"}
{"id": "train_23702", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "governance", "auditability", "security_gates"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1087713866939887d947f06be605f1b69be33bbb"}
{"id": "train_23703", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "tests_are_truth", "governance", "evaluation_metrics"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b5acd8ccc6190d420f139704c62226622694bba0"}
{"id": "train_23704", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "reproducibility", "security_gates", "tooling"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6b4e8c2688f75a57411ff6590f1075f5ff9f46bb"}
{"id": "train_23705", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "documentation", "evaluation_metrics", "ci_integration"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "315d4679b982904d79dd0f0371a9f8feba078702"}
{"id": "train_23706", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["auditability", "security_gates", "reproducibility", "tooling"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d4bba865387cf8093ee43a916f365844eaec7bef"}
{"id": "train_23707", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "auditability", "evaluation_metrics"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d9ce05fda1f0bd8d87b1ad59016d583307124232"}
{"id": "train_23708", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "reproducibility", "security_gates", "auditability"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e70d1e540ba2b6b449e70087020d28188f75e6fd"}
{"id": "train_23709", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "documentation", "security_gates"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "82e35e4d43cb17d0d973c363d9cc55202a50b963"}
{"id": "train_23710", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "reproducibility", "security_gates", "governance"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b639ec601af1d4b3310e0a0171ee0f5d6c451b2e"}
{"id": "train_23711", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "tooling", "reproducibility"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b970e5694ac346f3af910b118a6b8a77aa4db3f0"}
{"id": "train_23712", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "tests_are_truth", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fe719418fd9f3444d99aab71d7efc8ccdc5e1f89"}
{"id": "train_23713", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "reproducibility", "tooling", "ci_integration"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e74be75ce9f1fd9b62c708c5269947aaa4c8cec4"}
{"id": "train_23714", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "ci_integration", "governance", "auditability"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "82450652cced7a93d439e9853e54a2902d329e53"}
{"id": "train_23715", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "reproducibility", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1f15ab0e8f4db859209278ac672d2fafa985afde"}
{"id": "train_23716", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "documentation", "governance", "ci_integration"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "76a46615dcc12079f79e617b651d1a619c0611a4"}
{"id": "train_23717", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "auditability", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7e86364c0b25ce7dcf5f49eb523154450a5e2c2d"}
{"id": "train_23718", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "auditability", "tooling"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ee43ccaac1fbc6f6ccd119aed628638306a37703"}
{"id": "train_23719", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "tests_are_truth", "security_gates", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "83ec8782a39621696f53fce70587aef5834e90f5"}
{"id": "train_23720", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "reproducibility", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "106001e35537c95622013fc9a4f8245b24d1d9c3"}
{"id": "train_23721", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "tests_are_truth", "governance", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f008bfcc8eaaf356acf37dde58086b8e11c43345"}
{"id": "train_23722", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["governance", "security_gates", "reproducibility", "evaluation_metrics"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "64811dc13d998e0fcddbaba9eb599887f2809af0"}
{"id": "train_23723", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "repo_scale_reasoning", "reproducibility"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "abacfe585000362bd0103aedcda8418689483c93"}
{"id": "train_23724", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "auditability", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "752b1135d7b9e5998bad6002b2f8a11c9beac951"}
{"id": "train_23725", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "auditability", "cost_latency_tradeoffs", "governance"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9596786df1ef197704ff0b2a28015230bbf6b437"}
{"id": "train_23726", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "ci_integration", "security_gates", "documentation"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "82767d20dc5ebae1b1745ba1bf74c1d8c9b88e75"}
{"id": "train_23727", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "ci_integration", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bea23bd133244b16b8899bb89e01b7b02cc84366"}
{"id": "train_23728", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "security_gates", "governance", "tooling"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "769882db81fd6b4077cb55062ce4e6586600fb4c"}
{"id": "train_23729", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "security_gates", "evaluation_metrics", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "47d4e0d3533eded66ac4473bcc01e8468b67917a"}
{"id": "train_23730", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "tooling", "documentation"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3acd2f1743fc924339c3fd3e8bf90aebee90967e"}
{"id": "train_23731", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["documentation", "security_gates", "auditability", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "78f6d0adb083116c61802c9c3cd5f7da0074d802"}
{"id": "train_23732", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "tooling", "governance", "reproducibility"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "de453e3df04ab7a75af72fe36633879e3f8a0dbc"}
{"id": "train_23733", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "repo_scale_reasoning", "security_gates", "ci_integration"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "963702baf8d86ab6fb3adea400b82ef0cc2204bf"}
{"id": "train_23734", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "cost_latency_tradeoffs", "ci_integration", "auditability"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e531473fbb3c86823b6382cf519760019c14735a"}
{"id": "train_23735", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "tests_are_truth", "repo_scale_reasoning", "governance"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3b928ff425fe573f467167856be93300dced5cc1"}
{"id": "train_23736", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "tests_are_truth", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1b568ff0813a9303167163a8bb2103fe5771992e"}
{"id": "train_23737", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "documentation", "tooling", "auditability"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fe4641cf61da7a68ebe948213223163d32c6765c"}
{"id": "train_23738", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "auditability", "ci_integration", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0cb13ce8d15420f9eb018b0f4a5c70ade7a8b07d"}
{"id": "train_23739", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "security_gates", "evaluation_metrics", "governance"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e64b37d4c81548c502096e515c72c23e93ff45cb"}
{"id": "train_23740", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "evaluation_metrics", "ci_integration", "reproducibility"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "33b0416fcb421ffdb2b4b78f0d9036925b6bf88f"}
{"id": "train_23741", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "repo_scale_reasoning", "governance", "reproducibility"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "33b0aaded815c1054a41daf692f27427f481d7d5"}
{"id": "train_23742", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "documentation", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "48ba12da3d2855f46ed5477f4c87cded3af6864b"}
{"id": "train_23743", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "auditability", "reproducibility", "documentation"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "15b82e23b00e1f9dbd18fca1b39b4d25f06602a8"}
{"id": "train_23744", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "ci_integration", "auditability", "reproducibility"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f41fb8e25e7c7094552ebd178fbb7af807662924"}
{"id": "train_23745", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "cost_latency_tradeoffs", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "76a966eefba865e2fe349f6a3d10395a1bee4134"}
{"id": "train_23746", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "tooling", "auditability", "governance"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "73ba525496d7f5e7d21da1092dc4e0c5f33746e1"}
{"id": "train_23747", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "auditability", "cost_latency_tradeoffs", "governance"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "89f68247ec9d6245baf93506b4e029843c97f53d"}
{"id": "train_23748", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "ci_integration", "auditability", "security_gates"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e8085fe2b613a6702a8db8ac7f3ad0fb2fc3f493"}
{"id": "train_23749", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "documentation", "auditability", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1d83571eeaf64981ac38c62742db228e2e4cad4f"}
{"id": "train_23750", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "tests_are_truth", "evaluation_metrics", "security_gates"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9407d6690f4b127853c538338808f7db7702c32d"}
{"id": "train_23751", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "governance", "ci_integration", "documentation"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5d6f57fe762d49922768b34ed2ca1c0f871b8f38"}
{"id": "train_23752", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "tooling", "documentation", "evaluation_metrics"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "649b2f8c09c54d76b4c310b1ff90ed0feda24e26"}
{"id": "train_23753", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "ci_integration", "auditability", "documentation"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d121a8650573fdaa824bb5e3af0c900a6ced28aa"}
{"id": "train_23754", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "documentation", "tests_are_truth", "security_gates"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d5f4bc994779225cb34a9a6b3757769a6530a13d"}
{"id": "train_23755", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "documentation", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1bc4eb6c3862f829116c7c63e568fad0f7a7f113"}
{"id": "train_23756", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "governance", "reproducibility", "security_gates"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ff9260c4463042daba1689aef43c9133cc740918"}
{"id": "train_23757", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "governance", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0d93554c43c9958d025fb4b89cf0a3819bfc1d1d"}
{"id": "train_23758", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "auditability", "ci_integration", "tests_are_truth"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "25be2db15b6cc0a412d8466eb713c358f8d4c4d1"}
{"id": "train_23759", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "security_gates", "evaluation_metrics", "reproducibility"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "851a5469150e4e5bbbb2c7aa64679dc9c24748e0"}
{"id": "train_23760", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "security_gates", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7ec2600af037c7f4460a1e436cc2db76855c14c8"}
{"id": "train_23761", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "auditability", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "735e5b25be1b4953c525779778bb9314664a5bb1"}
{"id": "train_23762", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "auditability", "documentation", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6ed0551779f5236100d8c0e05128699eae668f8b"}
{"id": "train_23763", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "reproducibility", "security_gates"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "af8a8e03577041cf6ef0c1359f0a13a3b35051eb"}
{"id": "train_23764", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "reproducibility", "governance", "ci_integration"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "97a9d00133ebe0dd803f0a66e8f61b724abac2b8"}
{"id": "train_23765", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "security_gates", "tests_are_truth", "ci_integration"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bb9c3bf4132ba94ac313e96c00acb93e5bb8c999"}
{"id": "train_23766", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "security_gates", "repo_scale_reasoning", "auditability"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d59959c9669f2d5fd7fd4fbeb123400033a27af0"}
{"id": "train_23767", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "tests_are_truth", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e68a0da6665682b104a06fdfb380eaea0dcfa6da"}
{"id": "train_23768", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "governance", "repo_scale_reasoning", "auditability"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "737e346d2f0d738c7ce669c6ff2a8b080dcbdb91"}
{"id": "train_23769", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["governance", "documentation", "reproducibility", "auditability"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "88835deeb7b9cae958759be8f14d4083a7235d0d"}
{"id": "train_23770", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "governance", "reproducibility", "documentation"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d8aeada899da5817f73794b55aeecdc86e849604"}
{"id": "train_23771", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "tests_are_truth", "governance", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "38ce036abccfed75fe3abfcd8d07f032e5714d49"}
{"id": "train_23772", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "auditability", "tooling"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "aa9dfdf0f0287776b005c2abcd16143a5a80aee9"}
{"id": "train_23773", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "security_gates", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6f16beec1f142c9bef1f27c1503cc6658e869bc9"}
{"id": "train_23774", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "tooling", "documentation", "evaluation_metrics"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a9c14d0e0af4a537321c5807a7c58d676281ad01"}
{"id": "train_23775", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "governance", "documentation", "security_gates"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cf20e04a3bc81d228e121d0afb34e503dc1577b2"}
{"id": "train_23776", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "documentation", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "028587454e1e70fc03f16ebd05c31f831bede196"}
{"id": "train_23777", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "governance", "documentation", "security_gates"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "132df8d39b6a87d9d7d880646db1c6d1ae708ff2"}
{"id": "train_23778", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "auditability", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "196ef74efb445ba7d1b526b7328e0f5e82ab99ed"}
{"id": "train_23779", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "auditability", "ci_integration", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ac88249f2f98b8778649c93bd53dcbeef40954a8"}
{"id": "train_23780", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "ci_integration", "evaluation_metrics", "auditability"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c4f32b1d42669aab3fb1d8792c2db86d64367a10"}
{"id": "train_23781", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "auditability", "evaluation_metrics"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d074c17a0837fd1c7e0dd6d5d90d751d80e72577"}
{"id": "train_23782", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "ci_integration", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9a4a5048ca10e7709d6df5853b33ca894d2130cc"}
{"id": "train_23783", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "repo_scale_reasoning", "security_gates", "tooling"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ae64d3bd2ccc01633fc9ca4faa2483e9f94e0222"}
{"id": "train_23784", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "documentation", "tooling", "auditability"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "98a6080a000250dd420984dd037bf9cf93240718"}
{"id": "train_23785", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "documentation", "governance", "tooling"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "438e41be4677cd716440875351af88a91b5c36be"}
{"id": "train_23786", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "tooling", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "721731e2b9160beb4f329626808cd4fc88afccd5"}
{"id": "train_23787", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "tooling", "ci_integration", "tests_are_truth"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f9951c185210cb6e53c6bac087c9c6089a8ef37c"}
{"id": "train_23788", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "tests_are_truth", "tooling"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ced8577698dea8c61394b3a2c6220067a6257c5a"}
{"id": "train_23789", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "governance", "repo_scale_reasoning", "reproducibility"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d68a5ea985ef9cf2bacd00f165b5ebb51147108e"}
{"id": "train_23790", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "reproducibility", "cost_latency_tradeoffs", "repo_scale_reasoning"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "844b0545793c2aad3d87db563eaff94eaba829c1"}
{"id": "train_23791", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "tooling", "auditability", "reproducibility"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8ca6112a4953e4da5365c6503479db45c1255d40"}
{"id": "train_23792", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "security_gates", "documentation", "tooling"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "817e32321ddda7533f9e1ecc16d9f6b0b53a5e3a"}
{"id": "train_23793", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "governance", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "447e391e2740386379d43da6cc21cb92fad73aa2"}
{"id": "train_23794", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "auditability", "documentation", "evaluation_metrics"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3dc42da350b9142f03108aa554fb967f282ff9e0"}
{"id": "train_23795", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "documentation", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9efa1a9a772332e130f6d88ea946cfb361c36f36"}
{"id": "train_23796", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "cost_latency_tradeoffs", "security_gates", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3cdc46bc3cbdad20901ea8cd0d9f11a0cc8f8489"}
{"id": "train_23797", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "tests_are_truth", "security_gates"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c7f3fe1a1785fe935856c2b7e8483c8a727df1be"}
{"id": "train_23798", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "tooling", "documentation", "governance"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1de1bfb48e1dc6305d0b157d3d74996e35fe220c"}
{"id": "train_23799", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "documentation", "tests_are_truth"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cfcfc6cf531701550f98636e501dd8914c0ad7b0"}
{"id": "train_23800", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "tests_are_truth", "documentation", "tooling"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "22a63634bfe2d5f55e2843e7b5f4764188254f20"}
{"id": "train_23801", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "evaluation_metrics", "documentation", "auditability"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d6a85cccf6ab019a882cccfbd963171fa8e024c8"}
{"id": "train_23802", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "tests_are_truth", "ci_integration"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f88fe3cdb9e2c7fb2125e2db29162ea8c1f6753b"}
{"id": "train_23803", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "repo_scale_reasoning", "auditability", "documentation"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0d1b0f464d8d520dcaabefc6a4231e8f732a1e3d"}
{"id": "train_23804", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "tooling", "governance"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3ba22e18d977021279f8d6d34270c188b6b72522"}
{"id": "train_23805", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "security_gates", "auditability", "evaluation_metrics"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6b25694df0593a7e5ec51ba2292a8a45aedbfcaa"}
{"id": "train_23806", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "tests_are_truth", "auditability", "security_gates"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "076a6891bc2fb79069a57465291336d169ce75c0"}
{"id": "train_23807", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "reproducibility", "tooling", "governance"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "995f0dab2c7b6e09ed7994fbd1dcfe772e3d931d"}
{"id": "train_23808", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "reproducibility", "evaluation_metrics", "security_gates"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "89580e238df2034ab23fb67305313727dcdcf58c"}
{"id": "train_23809", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["documentation", "governance", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c7961e6ebb10b4c752fa48a14a47bec14d444119"}
{"id": "train_23810", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "auditability", "documentation", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "deca62b04555731a07efb421035237395e0b9cfc"}
{"id": "train_23811", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "tests_are_truth", "security_gates", "governance"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b20fa4b21fa02486eb7283b479854a251989b8bb"}
{"id": "train_23812", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "governance", "auditability"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2ef7765cb4ce83e7592b50c9e15b75931135c248"}
{"id": "train_23813", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "documentation", "auditability", "tests_are_truth"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ee63f681409f67ea3ac605de89d644a14e865432"}
{"id": "train_23814", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "tooling", "governance"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "044868caa0a35c6d219d7cc577af58bcf1a75f4f"}
{"id": "train_23815", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "repo_scale_reasoning", "tooling", "governance"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "468a738800c47e1c957c02db2a8b50918fdfdb4d"}
{"id": "train_23816", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "auditability", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a5cd725038e9be7984fdfb22ddf5f14a6e82bff9"}
{"id": "train_23817", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "tooling", "evaluation_metrics", "security_gates"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dd60819d44bf3f428b5736b6ef58fe9db1e9a981"}
{"id": "train_23818", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "repo_scale_reasoning", "security_gates", "ci_integration"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5ad0fd1c003cd10056eb669ea6e2c7fffe560a50"}
{"id": "train_23819", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "ci_integration", "documentation", "tests_are_truth"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e910d2c94f57dd63d51d92f7dd486c19a2e50f54"}
{"id": "train_23820", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "cost_latency_tradeoffs", "ci_integration", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "47a51a9c42854613216bc6eda849fe577698a407"}
{"id": "train_23821", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "cost_latency_tradeoffs", "ci_integration", "tests_are_truth"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ad91405fb04cfc82058f6f620d2dd8923f5efaa5"}
{"id": "train_23822", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "repo_scale_reasoning", "cost_latency_tradeoffs", "security_gates"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "03fb2111c0beefdf3aac5b3dd5e769c4280118cd"}
{"id": "train_23823", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "auditability", "security_gates", "ci_integration"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "499a9c207db59c5d920a6954b543967e3d6531f3"}
{"id": "train_23824", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "security_gates", "documentation", "governance"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c3b1a4c3ab6fbde57a887f65aaf767af191f28f1"}
{"id": "train_23825", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "governance", "cost_latency_tradeoffs", "security_gates"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4d6d386bb67bbb5f892930e615a5e003490b69d6"}
{"id": "train_23826", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "reproducibility", "governance", "repo_scale_reasoning"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "133ea289866d7fd69c0da57eb9c845347664c4d7"}
{"id": "train_23827", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "auditability", "evaluation_metrics", "governance"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "eab4c897ebf1d0cdd2a740e2be457a242b30c8d6"}
{"id": "train_23828", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "evaluation_metrics", "documentation", "governance"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4f2f36e3dee24967b0766a84ed77d50b91164b88"}
{"id": "train_23829", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "governance", "auditability", "reproducibility"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2ce8a6d730cbd25cb5abd104ff0c48e2c641bc6f"}
{"id": "train_23830", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "documentation", "security_gates", "tooling"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c4f229db37e1afb65304e191e73068599ba3dacd"}
{"id": "train_23831", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "tests_are_truth", "ci_integration", "reproducibility"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "29cb455cb46eee6e0a8174bdfa59e2c3e5bce77c"}
{"id": "train_23832", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "evaluation_metrics", "tests_are_truth"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "10b6dbed52a726cd7402bde72e8f5158b7d3b62b"}
{"id": "train_23833", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "repo_scale_reasoning", "evaluation_metrics", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a112c35499765c582d2e591467c83dce5398c648"}
{"id": "train_23834", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "auditability", "documentation"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c707f7d2f824d55c51775c93be8671804bbf0db9"}
{"id": "train_23835", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "reproducibility", "auditability"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "12867d7fdae49fc1fa54d238b6e95b0529788bf6"}
{"id": "train_23836", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "repo_scale_reasoning", "reproducibility", "tooling"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4f157b4345055db126aff29380cdb1b7eed1c0cd"}
{"id": "train_23837", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "repo_scale_reasoning", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c4d28bcfa7bc9b0226c32b9895f00b3def4a92d7"}
{"id": "train_23838", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "security_gates", "documentation", "ci_integration"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6d57071b8b90b1f27f223dae7ca2ef1c944e4e34"}
{"id": "train_23839", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "reproducibility", "tests_are_truth", "security_gates"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5b229649999b99b993bc6c09ea11c0de6936027d"}
{"id": "train_23840", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "security_gates", "ci_integration", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0a53248b6496e567ff96963f596bf192a03fe5d9"}
{"id": "train_23841", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "security_gates", "governance", "reproducibility"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "67739a46ce02aa93610412f5242d52a2053eb0ed"}
{"id": "train_23842", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "documentation", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a6ec48c4c10b8be47d6165e45830487e607dce36"}
{"id": "train_23843", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "evaluation_metrics", "auditability", "tests_are_truth"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "92c6319612594ea6bc781b563e5bee470f104878"}
{"id": "train_23844", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "governance", "ci_integration", "auditability"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9b59407bca6f9ce6d18d724f8533553b484d5436"}
{"id": "train_23845", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "auditability", "security_gates", "tooling"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "af102255028f89d052f4d8429e9784e3aa6eb932"}
{"id": "train_23846", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["auditability", "cost_latency_tradeoffs", "tooling", "ci_integration"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "118c5013e7886f32c7f2f82977ca39d14239da49"}
{"id": "train_23847", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "tests_are_truth", "ci_integration"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dee4bc85af11249ce86c90a3d6112ab529f65bf0"}
{"id": "train_23848", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["governance", "evaluation_metrics", "reproducibility", "tooling"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "48e91fd54557a76e782766945ce6400965e5f833"}
{"id": "train_23849", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "ci_integration", "tooling", "tests_are_truth"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9f4e691b1c8acd9bf668c2adc858a44b6eee7766"}
{"id": "train_23850", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "governance", "documentation", "security_gates"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ae5877110e6c9df710ae022e1a460b797ed607f8"}
{"id": "train_23851", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "documentation", "security_gates", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "12a124511614ccc18202191130a5ad2123f169d9"}
{"id": "train_23852", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "documentation", "tooling"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5d3e7461e35d01a17637fb6e8257748c9ca61688"}
{"id": "train_23853", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "tooling", "ci_integration"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f8a192a826aac93cd936e545d6c80b9ca0508ded"}
{"id": "train_23854", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "documentation", "security_gates", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0402ea2401676b6c4177a4c0170a096b191732de"}
{"id": "train_23855", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "tests_are_truth", "auditability", "tooling"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a62cc853dbedd770222b05aa37d7da84cc34be68"}
{"id": "train_23856", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "governance", "repo_scale_reasoning", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d06b78a076fb158cd966d868aff7f0ee3a677508"}
{"id": "train_23857", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "tooling", "security_gates", "documentation"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "69887d356b8a20d0f90f5680baa80e3289c8036e"}
{"id": "train_23858", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "ci_integration", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d4f55c11c0806b2fc4f029771505c3d18f53fdde"}
{"id": "train_23859", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "governance", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4b485cf1af5058024b74ee1840ac0853d7e3ebad"}
{"id": "train_23860", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "evaluation_metrics", "tooling"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6e3017df39fba547d97bd3c6668f954455f62d17"}
{"id": "train_23861", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "tests_are_truth", "evaluation_metrics", "security_gates"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8b59d1bc448e936582d93e2a782a68cdbc4e3068"}
{"id": "train_23862", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["governance", "auditability", "reproducibility", "security_gates"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ff167e2dd89774f30df63247c7a06cd505951765"}
{"id": "train_23863", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "auditability", "evaluation_metrics", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "170cf5766a6124910b78572e7b89ed7834773c4a"}
{"id": "train_23864", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "reproducibility", "security_gates", "ci_integration"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "daa6fc42a7b7303c918f616bbd3ade7841b99b63"}
{"id": "train_23865", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "security_gates", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0e7278895535c4ff6bba2f816a56bfdb640fe930"}
{"id": "train_23866", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "tests_are_truth", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b3f215c4b8480684a07bc3b53157ca8f72178a2c"}
{"id": "train_23867", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "ci_integration", "reproducibility", "evaluation_metrics"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ed50f9ca036d3dddc27f5dda796fe5179d3f272f"}
{"id": "train_23868", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["governance", "repo_scale_reasoning", "auditability", "reproducibility"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3177de63b9e645fbc64b4b80b53666f3ed86e51a"}
{"id": "train_23869", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "documentation", "auditability", "tests_are_truth"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "714ffbf590fea94c608e25b820f62ded35af5f48"}
{"id": "train_23870", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "security_gates", "governance", "tooling"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "90434df9c9fc4f4a4c2ba1952f3a9288f455ef41"}
{"id": "train_23871", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "documentation", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "607e455175716d5e7e33769416f7f5038896e398"}
{"id": "train_23872", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "cost_latency_tradeoffs", "governance", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f1b6afec66f40f712ec729bf494666ed0462bbc4"}
{"id": "train_23873", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "tooling", "reproducibility", "security_gates"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0eabf520cea127a29430e5f4e385f2c8d6b4f193"}
{"id": "train_23874", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "tooling", "documentation", "tests_are_truth"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a78b79475147dc52f5e6e3853e091f99a01e9a61"}
{"id": "train_23875", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "tests_are_truth", "tooling", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b493fb340ccfc2605f8c7d0cc803318dc05c212f"}
{"id": "train_23876", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "security_gates", "ci_integration", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "60069090cb0d4ad3f4368bed9929e4eaa7a1cba6"}
{"id": "train_23877", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "governance", "auditability", "reproducibility"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3c6e9e29341b971a75839d4c8dffb3ca2f6ae4e3"}
{"id": "train_23878", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "security_gates", "auditability", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3588e24b4b7f40ce7cb59929296aafb6107727d3"}
{"id": "train_23879", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "documentation", "tooling", "ci_integration"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "caa222ce71aadd887afe456feff5a2f3264c49a7"}
{"id": "train_23880", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "ci_integration", "tooling"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3181cfda7500b7ee8a8ec6fc53f2c67ee912dfab"}
{"id": "train_23881", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "documentation", "governance"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "632746ef806ea48679127fb3ef0051e32a3091bc"}
{"id": "train_23882", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "documentation", "tooling"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d89fabf4e57f1adf45d091a12be00c9e23d70812"}
{"id": "train_23883", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "governance", "documentation"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "31314d603f156b1e52b49b1b2880952a2cee2ec2"}
{"id": "train_23884", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "documentation", "evaluation_metrics", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6348e74c5fb9e8f81de48b190b5b004de7b26106"}
{"id": "train_23885", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "auditability", "security_gates", "tooling"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9d75dd0fae0e63704a3807d112db21581443ebed"}
{"id": "train_23886", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "ci_integration", "repo_scale_reasoning", "tooling"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "72bf759fb93ac00db06437933147be5f874353ff"}
{"id": "train_23887", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "repo_scale_reasoning", "reproducibility", "tooling"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "db01d5574ceaef5cfe227783d21f14eb16a5d056"}
{"id": "train_23888", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "auditability", "tooling"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "35a35a2b6595a061e12a2f0d5825d4ab7e4c2127"}
{"id": "train_23889", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "ci_integration", "documentation", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3c3861761b82a5e17b359f51af8da33caa754893"}
{"id": "train_23890", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "ci_integration", "auditability"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e268c099a953ed1fd9d1bb9517d0829ed0052061"}
{"id": "train_23891", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "auditability", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6c2eb434ccf16f13a55c2a3742c34e32aaef8cdf"}
{"id": "train_23892", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "governance", "documentation", "repo_scale_reasoning"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e4b351df7b9c3636229ec5cc574a8e7c759bc175"}
{"id": "train_23893", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "governance", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a0bf0c5bcc7d2472ebaefd1b811dbe3082321637"}
{"id": "train_23894", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "auditability", "reproducibility", "ci_integration"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b3f810931fc4e08787316c898fdd80371aa0b58c"}
{"id": "train_23895", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "cost_latency_tradeoffs", "reproducibility", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2fb956d8965af5d9b9d7ef3492c606fcb2486136"}
{"id": "train_23896", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "repo_scale_reasoning", "auditability", "reproducibility"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7bd26117151e31e6183df27415d5b9d9a1418faa"}
{"id": "train_23897", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "ci_integration", "tooling", "governance"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cb769ad27bbf7b27b918f24072815ae69924ca63"}
{"id": "train_23898", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "tooling", "repo_scale_reasoning", "auditability"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "148c68bddcaea4d921c8ff883ea94091b7fbc521"}
{"id": "train_23899", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "security_gates", "documentation", "reproducibility"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "60069090cb0d4ad3f4368bed9929e4eaa7a1cba6"}
{"id": "train_23900", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "reproducibility", "evaluation_metrics", "auditability"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "873c174b40a5cfe5bfc0c930c8dde9a1c6dbb405"}
{"id": "train_23901", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "documentation", "security_gates", "auditability"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e032b2d3de634e79492d713cfb0f11c840b64984"}
{"id": "train_23902", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "governance", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8c5e16edd80a760443128b577ebe6b9b221560bc"}
{"id": "train_23903", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "tooling", "security_gates", "reproducibility"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "80a131a253d4164ba12ebe4da03284a300026617"}
{"id": "train_23904", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "reproducibility", "evaluation_metrics", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "88e85bbe324962dfb5782af7dad49853d2c5bd7e"}
{"id": "train_23905", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "governance", "auditability", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "59ee8b87d26e6796072404a0625d87a0aae38c7c"}
{"id": "train_23906", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "ci_integration", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3885cf90fe4d450959140992ae6584ace14cd7f1"}
{"id": "train_23907", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "tests_are_truth", "security_gates"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cf8cbe66b421a12f27bebf637ea77b2067055a74"}
{"id": "train_23908", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "governance", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "805b8154faddbf1963a15a64433bf336d86e4ec7"}
{"id": "train_23909", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "tooling", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b2b50fa3331b2f507d5a6cdb177f52ae0634cc13"}
{"id": "train_23910", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "governance", "tooling", "tests_are_truth"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "95d630d668d0debba33a1b618ee9f71731814562"}
{"id": "train_23911", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "security_gates", "auditability", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fe671a5786364c4209fede8f501aabb331fc8451"}
{"id": "train_23912", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "repo_scale_reasoning", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "209a44ce144613b436cb53cc7ccaa73ce072c8ec"}
{"id": "train_23913", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "documentation", "evaluation_metrics", "governance"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8adf799b4f54b181061f654e6dd854c6daa9f897"}
{"id": "train_23914", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "security_gates", "auditability", "ci_integration"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "40073d5143979aa41b61c2e0c77403af17e243ea"}
{"id": "train_23915", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "security_gates", "ci_integration", "documentation"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5d3e7461e35d01a17637fb6e8257748c9ca61688"}
{"id": "train_23916", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "repo_scale_reasoning", "tests_are_truth", "security_gates"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dd841c9ebb2d22a542541958735f4fc480d80b7b"}
{"id": "train_23917", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["documentation", "repo_scale_reasoning", "security_gates", "reproducibility"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a397bbd66c73b683f25e10c49e6af49741074ceb"}
{"id": "train_23918", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "documentation", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "47b39ab29f9afcec2d53f12282ea853f116f5912"}
{"id": "train_23919", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "governance", "tooling", "reproducibility"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e817291205488dce97e10aad5645bb2c6af5a436"}
{"id": "train_23920", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "tooling", "ci_integration", "security_gates"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c0480fcc8253003b734521660d018c5992ee00c1"}
{"id": "train_23921", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["governance", "repo_scale_reasoning", "tooling", "reproducibility"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0fcfaa19e5c24e9741347b27e91adaa4f066093b"}
{"id": "train_23922", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "documentation", "cost_latency_tradeoffs", "governance"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e2e608dac3ce1b5289b795662b529301ca3b9f42"}
{"id": "train_23923", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "tooling", "security_gates"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4af52c3fd93010ec005dd5f95910efa89981cbe1"}
{"id": "train_23924", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "tests_are_truth", "repo_scale_reasoning", "auditability"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dc735220db6fe5787d570ddd1d27136572876bf9"}
{"id": "train_23925", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "evaluation_metrics", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fde16d633be6e66eb62077491b5e7450594e53f7"}
{"id": "train_23926", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "evaluation_metrics", "ci_integration"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c0ebe1699ba47bc53a6f730a2663fb609ceee97b"}
{"id": "train_23927", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "ci_integration", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9816a15f05ac1c78be8e67a4e1cece98850a3bf8"}
{"id": "train_23928", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "security_gates", "documentation", "tooling"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f973a22c3a0e9e95fc817f9a1d748237114c5da2"}
{"id": "train_23929", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "tests_are_truth", "documentation", "evaluation_metrics"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "30e2935af6c5158ffcfff73d665fcb8141adf404"}
{"id": "train_23930", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "auditability", "security_gates"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8b8546d15cab5fb020b679f6e666bf96836def77"}
{"id": "train_23931", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "tooling", "ci_integration", "evaluation_metrics"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "256c5e34dddcc07f3176ed1acb32f8bd61a667c5"}
{"id": "train_23932", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "security_gates", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "75d5c5ff9d16fcd47bee6adcf58dea3ec3480b16"}
{"id": "train_23933", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "governance", "evaluation_metrics", "reproducibility"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "81ccac023bad1d01c0365ec4937a32e3bb4c7551"}
{"id": "train_23934", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "reproducibility", "tooling"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "efdf60299a35728342211a038dbd47d1e014f8db"}
{"id": "train_23935", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "security_gates", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ca94b50467dcc31e257bbf61764b7ec3e6a9e916"}
{"id": "train_23936", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "repo_scale_reasoning", "documentation", "tests_are_truth"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6ab5f74f7bc4d955678c20f7da27373ed8c0da15"}
{"id": "train_23937", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "tests_are_truth", "governance", "evaluation_metrics"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d504d7474dfa4ca7343e2d1ced550caa9647c4b6"}
{"id": "train_23938", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4db337c377d8ca6ee278b18fec98ea1bae97c79d"}
{"id": "train_23939", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "tooling", "auditability", "documentation"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "52c48b33267de16d2f2648633d8bc3d5ab9dec35"}
{"id": "train_23940", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "evaluation_metrics", "governance", "reproducibility"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "96059990a56167f4bc79fdf3b82046b26c99c212"}
{"id": "train_23941", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "tooling", "security_gates", "auditability"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7531cc499bd8b370bfce612cb4f9c308aa4182b6"}
{"id": "train_23942", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "repo_scale_reasoning", "governance"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d8762852519ec5ff27bda31e9187493f6916650e"}
{"id": "train_23943", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "documentation", "governance", "evaluation_metrics"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f0faf7ddb1d42bd17728f01a2f844ec0180639d2"}
{"id": "train_23944", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "tests_are_truth", "documentation", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b1ad77e2e9e682caccd67be6a8c0a7486f5b8a09"}
{"id": "train_23945", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "auditability", "tests_are_truth", "governance"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d9dbaf3ee9572b3f2e9d33453c0e7df3a3923af1"}
{"id": "train_23946", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["governance", "tooling", "auditability", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d34176664be2afe15fcce3861a6becb8e6ef30db"}
{"id": "train_23947", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "ci_integration", "security_gates", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1866f00525595df28b06de51676688d8472f39b1"}
{"id": "train_23948", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "ci_integration", "tests_are_truth"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e8b3bc670372e613b4f37848d7a6f312e9b36c34"}
{"id": "train_23949", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4a11b587ec59c7b97f2d271e86a6bea25105fd24"}
{"id": "train_23950", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "reproducibility", "repo_scale_reasoning", "documentation"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fea339960751ce506d23613f9daf12fd253bbd84"}
{"id": "train_23951", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "ci_integration", "security_gates", "evaluation_metrics"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9b56a2a40b1a80a88e2442734354abb2b18416de"}
{"id": "train_23952", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "auditability", "ci_integration", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3892395ce89d09a1e7c5010b4cbc6686102b3ec5"}
{"id": "train_23953", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "tooling", "auditability", "reproducibility"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "54bcd8e3b42da83f28bf0bb492be083698f9320f"}
{"id": "train_23954", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "tooling", "auditability"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "edac475c79f36bf651035836c7712d9ea1ec4ad9"}
{"id": "train_23955", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "tooling", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fe948ecd77d0c3dfc45baf70296a83b853e78278"}
{"id": "train_23956", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "reproducibility", "auditability", "ci_integration"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4d8e6a164687cc83ccc1c4b929f9cabc7b26c558"}
{"id": "train_23957", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "evaluation_metrics", "tooling", "auditability"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "41c1c9781c9c55f0c5e57f8b105469ce45c9a347"}
{"id": "train_23958", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "security_gates", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9d3d89c956115ebe8b045d3f97d3c29f9a3188a4"}
{"id": "train_23959", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "repo_scale_reasoning", "auditability", "governance"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ba914c3b2dcb00a5b232282865bad9166307df4c"}
{"id": "train_23960", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "documentation", "cost_latency_tradeoffs", "security_gates"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "be6f454fd9b0563298ebbd7b0b8ac1202ff2f8dc"}
{"id": "train_23961", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "ci_integration", "documentation", "reproducibility"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2672c5a5e34dd2dba0e03148227a0f5cfffa134c"}
{"id": "train_23962", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "auditability", "documentation", "governance"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "058fe2ec8bcd61067d2f13ec42e5b01ffb9415f1"}
{"id": "train_23963", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "cost_latency_tradeoffs", "repo_scale_reasoning", "documentation"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "708fd225971279ecd5c92850b63ef124b95baf6c"}
{"id": "train_23964", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["governance", "security_gates", "evaluation_metrics", "auditability"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6824960fdd48b57a08e76ed6f001ea39f8e349b8"}
{"id": "train_23965", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["auditability", "repo_scale_reasoning", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3459d317f3d4dcbe71806cfc80fad42024325853"}
{"id": "train_23966", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "cost_latency_tradeoffs", "evaluation_metrics", "ci_integration"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1e5d7e046d09d9f318d38bc420859b3a8a5bcc62"}
{"id": "train_23967", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "evaluation_metrics", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2c2410119e2c47ea6eebfa3b203d33ef5d1af410"}
{"id": "train_23968", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["governance", "evaluation_metrics", "repo_scale_reasoning", "auditability"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f35ff22fceed910dfa0cf77e5ca24ffafbb98c5b"}
{"id": "train_23969", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "security_gates", "reproducibility", "tests_are_truth"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "afd736b255b5d3a65ad5444160abb1500914624e"}
{"id": "train_23970", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "cost_latency_tradeoffs", "evaluation_metrics", "documentation"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8830313816679601f3e405e469cfe5fcde35e5b7"}
{"id": "train_23971", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["governance", "documentation", "tests_are_truth", "tooling"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0c2290ca9218a8d3f6adde5c5ac2dafb053b9f9f"}
{"id": "train_23972", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "evaluation_metrics", "repo_scale_reasoning", "documentation"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "78ebf1895ff2e7cdcf48fc91649fc946c3e3b6e1"}
{"id": "train_23973", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "security_gates", "documentation", "auditability"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "850c9df216639189b12d76a62c595b0b7cbd68c0"}
{"id": "train_23974", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "security_gates", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ae9d24a9fe4e35b17afef23d141055171863ed2e"}
{"id": "train_23975", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["governance", "repo_scale_reasoning", "evaluation_metrics", "tests_are_truth"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6e9e5742baa5c10685d2a91bf8c46b571deee2e6"}
{"id": "train_23976", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "governance", "auditability", "evaluation_metrics"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c767c4a69e50d80568e58468079364d10bd62a60"}
{"id": "train_23977", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "cost_latency_tradeoffs", "reproducibility", "tests_are_truth"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8f3b82c3b46ce5151983fdf71c7c46aa7b245ed4"}
{"id": "train_23978", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "tooling", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "262d13204dd129cad06e8459d2dd626633d338fe"}
{"id": "train_23979", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["governance", "security_gates", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dd9407614dae33fab3804ca33ff03943c52acfd4"}
{"id": "train_23980", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "ci_integration", "tooling", "governance"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "aee61fba226a8562d98e0a2592eafa8383b69247"}
{"id": "train_23981", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "cost_latency_tradeoffs", "reproducibility", "governance"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2d9a278f7ae532c247286c0075f2da7af3c0f6c6"}
{"id": "train_23982", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "documentation", "security_gates", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2e42b09f292af5cb8bf8fddb0e8d1045ab45c2cd"}
{"id": "train_23983", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "ci_integration", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ea6b796b376c413588ebcf19fb1a17d36f0e9b8e"}
{"id": "train_23984", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "security_gates", "evaluation_metrics", "auditability"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "444f735101d6ce7c1db75d7e8afa3fbbbf08f3d1"}
{"id": "train_23985", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "tooling", "repo_scale_reasoning", "governance"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "15b5ae78aff40c76c3f3982906281409dd0d31f4"}
{"id": "train_23986", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "governance", "tests_are_truth", "reproducibility"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0aa2072913acb43b1f1888ee5eecefb54851414f"}
{"id": "train_23987", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "documentation", "reproducibility", "governance"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e3e0d36bc2d25026177bbfc402521a082a702656"}
{"id": "train_23988", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "auditability", "tests_are_truth"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2ca73aa0ff6ec5b240b1048aefdda2c6a77265fd"}
{"id": "train_23989", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "ci_integration", "reproducibility", "tooling"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "794dfa5c172fcd10966bee7f8d18724967e44974"}
{"id": "train_23990", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "documentation", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9b424b42b1b2fda4c9bc801a8548a6e323b06f03"}
{"id": "train_23991", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "tooling", "security_gates", "evaluation_metrics"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "68100dafaf01a381b498dfd620a1882b4d200786"}
{"id": "train_23992", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "auditability", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "775c46b625bf4e7c7c10ee4ddd9ea29ff14fd437"}
{"id": "train_23993", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "tooling", "governance"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4001de3bd5886beb6542bc04a851e909331c4f30"}
{"id": "train_23994", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "ci_integration", "reproducibility", "governance"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "18c3a39c067612dd1771629f70767e16e3949f47"}
{"id": "train_23995", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "security_gates", "evaluation_metrics", "tests_are_truth"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "be62511b2fe1b17704834cfd463e96c5e536fd8b"}
{"id": "train_23996", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "evaluation_metrics", "repo_scale_reasoning", "documentation"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9b64891b0482db718510eec8a8c956978cc297f7"}
{"id": "train_23997", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "auditability", "ci_integration", "governance"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "abb3da8d765df5f6b947c70baef2d7bdae47f205"}
{"id": "train_23998", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "repo_scale_reasoning", "reproducibility", "auditability"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3d201b6ff079f1baf371487c520052dcb086ae96"}
{"id": "train_23999", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "tooling", "ci_integration", "governance"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6271308fcac03432bffbcd055fe8ac09ad9db738"}
{"id": "train_24000", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "tests_are_truth", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "78603d76027931318ba5ba5931653827fcb35cde"}
{"id": "train_24001", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "documentation", "ci_integration", "tooling"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c5ff53afe4a9a2f23985eb6f51620b6233f2a8c9"}
{"id": "train_24002", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "reproducibility", "tests_are_truth", "tooling"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "459fa43b6486e1e149c796025265d889b8945fbf"}
{"id": "train_24003", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["documentation", "tooling", "evaluation_metrics", "ci_integration"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "75cd1c39f44127b101ba621e8ec9d225e6580fa8"}
{"id": "train_24004", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "auditability", "governance", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "438e41be4677cd716440875351af88a91b5c36be"}
{"id": "train_24005", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "evaluation_metrics", "tooling", "auditability"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fa96aae08ae0cac789464d90128d6450a7989b07"}
{"id": "train_24006", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "cost_latency_tradeoffs", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0d9da5af97d26d8183b087e4e7a94a7e7954fc3f"}
{"id": "train_24007", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "evaluation_metrics", "documentation", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "976d21d3ee75ce7f17c61255223052609f0c4326"}
{"id": "train_24008", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "tooling", "governance", "tests_are_truth"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "293534e855c285f83d0dd6887b8ea828b7079bed"}
{"id": "train_24009", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "ci_integration", "tooling", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e5914219d38a77dc96c9cb0b98c9df8a30829fae"}
{"id": "train_24010", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "reproducibility", "evaluation_metrics", "auditability"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "541bd0783867762e8dd2dacdc3059697424668ad"}
{"id": "train_24011", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "reproducibility", "security_gates", "documentation"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a74c69ed91fc9d8dfafd886a26eeec6c868b92f8"}
{"id": "train_24012", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "governance", "security_gates", "ci_integration"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "aac2ac08171e5392c7f006dca3cffc87e92512df"}
{"id": "train_24013", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "documentation", "tooling", "security_gates"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "492ba6d8500e00f03fc3dd08024f24f7423abc9a"}
{"id": "train_24014", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "evaluation_metrics", "security_gates", "tests_are_truth"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "19cb6bb1d3ccdf68403668b0c09a17ffdfc019ea"}
{"id": "train_24015", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "evaluation_metrics", "governance", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6eeb467590c7e53e4d99460eee438f3a12c9bc6f"}
{"id": "train_24016", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "auditability", "documentation", "evaluation_metrics"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d574114e6989128b197f491ede20a60206672252"}
{"id": "train_24017", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "tests_are_truth", "repo_scale_reasoning", "documentation"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "58685895c5a02dcd59b89fca86c3cff0379d428c"}
{"id": "train_24018", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "documentation", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "eb8c841e3d25d559d2897c48f4adcfe372108999"}
{"id": "train_24019", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "reproducibility", "documentation", "ci_integration"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6076aeb76c4502e7f94f6bfe6ac10bb099641d53"}
{"id": "train_24020", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "evaluation_metrics", "tests_are_truth"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fae0056bd2b726f001aa7bc94b2ac72c5614e3cc"}
{"id": "train_24021", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "auditability", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "00446218973c848d436207d8cb9bfb9c32f185eb"}
{"id": "train_24022", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "ci_integration", "evaluation_metrics"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9fd0ab7592067a370e546a13c8e0da058a00c648"}
{"id": "train_24023", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "tooling", "evaluation_metrics", "auditability"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "306afdd6977e3724e79e72d13f7527cd8ca19f8c"}
{"id": "train_24024", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "evaluation_metrics", "reproducibility", "ci_integration"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "496d95f76d8a4412af0d1f426e8a7b3d08c7e722"}
{"id": "train_24025", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "reproducibility", "repo_scale_reasoning", "auditability"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "03ef7abe2fe5dd17ee94d77e61e177d3220594a5"}
{"id": "train_24026", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "ci_integration", "cost_latency_tradeoffs", "governance"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "36022369d9feab22c1be44bb5e5c890522499ad6"}
{"id": "train_24027", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "security_gates", "repo_scale_reasoning", "reproducibility"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "422ce17eb9064485da85ec9d2b1a7758fee10e71"}
{"id": "train_24028", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "tests_are_truth", "governance", "security_gates"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "440aa4a0639c7c385f9387bfefe0133f3e68bff6"}
{"id": "train_24029", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "governance", "security_gates", "documentation"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ff4bc321ab167ef2b4c4136ccd07756f6b3a9462"}
{"id": "train_24030", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "security_gates", "ci_integration"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4272a0e4b82bdea55b5fb632f091fc656af01925"}
{"id": "train_24031", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "security_gates", "governance", "ci_integration"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "59647fb77e1db8ebf504a56fea144566536fb4bc"}
{"id": "train_24032", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "documentation", "governance", "auditability"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5c3e0f67c2dc4315a14a41a6bc51059391ca63ba"}
{"id": "train_24033", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "evaluation_metrics", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fd9869b51ec4d1f39aad5c749463f319902d5f64"}
{"id": "train_24034", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "security_gates", "tests_are_truth", "ci_integration"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a28580e3db23259bf9532f6a18a0ee599328f7f3"}
{"id": "train_24035", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "documentation", "tooling", "ci_integration"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3b631f1e79c51c1b23cf15b678c2224bc25da9f3"}
{"id": "train_24036", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "ci_integration", "tooling", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "06e9f358ff1604d5396e0b4df369bfbb38b22f81"}
{"id": "train_24037", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "reproducibility", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bec6b4985734b85e1d243cf915bb3c35889704e0"}
{"id": "train_24038", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "tooling", "cost_latency_tradeoffs", "security_gates"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "34910bbc31a666a9e7023d780bbee03f1157b2ff"}
{"id": "train_24039", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "evaluation_metrics", "governance", "auditability"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d182c457c3339bfd72c23a099ddf6b51ce3dabe9"}
{"id": "train_24040", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "reproducibility", "evaluation_metrics", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "872effa4de644390e02659809f67d30004f17ce2"}
{"id": "train_24041", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "tooling", "documentation"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8f52e68d77a996c568f64dacb3cc71df0e58f3fb"}
{"id": "train_24042", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "ci_integration", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "201083372ed4e03c26fefa4d1f906d465c140a52"}
{"id": "train_24043", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "repo_scale_reasoning", "auditability", "tests_are_truth"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "20feb372dc986d302f2812f0ce43a05e3031f5f8"}
{"id": "train_24044", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "documentation", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fd8126c1c18080eff2b63c25bd429068e1c0b3d7"}
{"id": "train_24045", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "governance", "reproducibility", "security_gates"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8408c38f3276a5e81b338690d90f1b587341f9ef"}
{"id": "train_24046", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "auditability", "governance", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d1bdf9b082f2f8e89439c8fd3813b37f36c5ee2b"}
{"id": "train_24047", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "auditability", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1424c4013100c480e981172a40cb52b6f7376f79"}
{"id": "train_24048", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "ci_integration", "reproducibility", "evaluation_metrics"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2fd06fc4744c8425ba351b80c818590e426c1f31"}
{"id": "train_24049", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "security_gates", "repo_scale_reasoning", "reproducibility"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3add3d6fb41910e951d117dbd74ef92a33a1041f"}
{"id": "train_24050", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "tooling", "governance", "evaluation_metrics"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "09a79990b2e8019383c5838350ed2c2e79a8648b"}
{"id": "train_24051", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "governance", "repo_scale_reasoning", "auditability"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "64a82012a6fca645fe87c92327e6ca3f334d8647"}
{"id": "train_24052", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "evaluation_metrics", "reproducibility", "governance"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f549eea282e3aac57e974d76847ce021cecde80c"}
{"id": "train_24053", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "governance", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "452a0561b856ff4c2b4f207fc8c8bd451f4092ed"}
{"id": "train_24054", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "ci_integration", "tests_are_truth"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f0e3590969e48b89cf6e3f18c897d91a490ce58f"}
{"id": "train_24055", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "evaluation_metrics", "documentation", "ci_integration"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b307c2c113faf755986318b1803efcbe52f5498e"}
{"id": "train_24056", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "cost_latency_tradeoffs", "tests_are_truth", "reproducibility"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d5f950bf499cc77e3efb651ea185ffadf6429eeb"}
{"id": "train_24057", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "auditability", "reproducibility", "tests_are_truth"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7df8d565164079cc75d4baf34fb027d998147f78"}
{"id": "train_24058", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "evaluation_metrics", "documentation", "tests_are_truth"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "304db5a7c62fe821b5d309dd872b4247ee70fc25"}
{"id": "train_24059", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "ci_integration", "security_gates", "tooling"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4469fad7a180369c3fa07132ba473bd78028c1a8"}
{"id": "train_24060", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "tooling", "evaluation_metrics", "tests_are_truth"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "698253205af94eeb763b5276ef4021507eb0ab09"}
{"id": "train_24061", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "reproducibility", "tests_are_truth", "documentation"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a0bf0c5bcc7d2472ebaefd1b811dbe3082321637"}
{"id": "train_24062", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "repo_scale_reasoning", "governance", "tests_are_truth"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "25cf435c7a87fb3894571515f7789ea9e5bc4f25"}
{"id": "train_24063", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "evaluation_metrics", "reproducibility", "governance"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ac9f014a33d759bae7403214caa385905eae1a71"}
{"id": "train_24064", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "evaluation_metrics", "governance", "security_gates"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "22846a13ba9e637943e456caeb408bd67ec0e3a8"}
{"id": "train_24065", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "tooling", "repo_scale_reasoning", "security_gates"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6cf2f3cc7e1d7521b5dd1e473fc260aa8ab8e48d"}
{"id": "train_24066", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "ci_integration", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c94b16ce36e1f579d4fff16241ec784e8dd552fd"}
{"id": "train_24067", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "tooling", "ci_integration", "documentation"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fd608f4bc8d3799f3a1151e01489f225a55ac55d"}
{"id": "train_24068", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "documentation", "evaluation_metrics", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1c6ed7af229fbea693d7dcb0eb3f98cf2b03a909"}
{"id": "train_24069", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "ci_integration", "evaluation_metrics", "governance"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "20eae69fccc88d7fed4961de78ebfa17b15adf0b"}
{"id": "train_24070", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "tests_are_truth", "governance", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fcce28b85c0bccfe4b2539f327ba86d98dd3aea4"}
{"id": "train_24071", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "repo_scale_reasoning", "reproducibility", "security_gates"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "966ad3bd20415fcfd8d4d44f0ec8122bfffd0bdd"}
{"id": "train_24072", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "documentation", "ci_integration", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "28de5dab2c841a63dab75ed17e723d29fb11dbbb"}
{"id": "train_24073", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "tooling", "documentation", "repo_scale_reasoning"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ff2ede8739ca7cc0059cc65860d8f25d52791d9c"}
{"id": "train_24074", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "security_gates", "ci_integration", "tooling"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "caf01ef2dfe51ca537116a188d195c9c96bf09ee"}
{"id": "train_24075", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "auditability", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c03a9610090220013a3928ff5b5e46df816f74b1"}
{"id": "train_24076", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "cost_latency_tradeoffs", "ci_integration", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "82ff61824d05ec68609a38016e9e4dcb004c47f1"}
{"id": "train_24077", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "documentation", "auditability", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1dc1ebc017da68efe6342af346b7b52d2bf23fca"}
{"id": "train_24078", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "governance", "ci_integration", "security_gates"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0ef66801e2391ba6d7a9d0bc1eb93625d38e48c1"}
{"id": "train_24079", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "tests_are_truth", "reproducibility"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "14b2e488b0c2388d052a9ec2ff9c531a1a774997"}
{"id": "train_24080", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "ci_integration", "repo_scale_reasoning", "tooling"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "38d5d516aa3340f21f8699f40dbd57839c5c3eae"}
{"id": "train_24081", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "evaluation_metrics", "security_gates", "auditability"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "92dd7a24e3a679d722c4fbac469c33a49b69eab5"}
{"id": "train_24082", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "tooling", "documentation", "auditability"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8ce6a746c534eb429379dae811ff90339411f6fe"}
{"id": "train_24083", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "security_gates", "tooling", "reproducibility"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b1387fbec1c8d21a3d973a3a48876e38dd93c039"}
{"id": "train_24084", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "tooling", "documentation", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "963a290a009c8df2031e860a5991793a9d1991a5"}
{"id": "train_24085", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "reproducibility", "governance", "documentation"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a5a6627670778bf5be6c0dd942070f310a1f27bd"}
{"id": "train_24086", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "ci_integration", "governance", "tooling"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "13e24033c7358a61fb54ccff8340dde8cc07871a"}
{"id": "train_24087", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "documentation", "ci_integration", "security_gates"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "77572388ecc2b8b935f3eefe5f23865bbd800bcf"}
{"id": "train_24088", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "tooling", "evaluation_metrics", "tests_are_truth"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ed312d31d2955b82a86d2cda769a08c499772bfd"}
{"id": "train_24089", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "tooling", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2bccb6f534aeef2457ac0388ef499f68b289c470"}
{"id": "train_24090", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["auditability", "ci_integration", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e9b59f99a0ebaae597b373709bc7c0f03a1cc3f0"}
{"id": "train_24091", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e6b99d0452e381e84750d7263bde3a4d93db7695"}
{"id": "train_24092", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "auditability", "repo_scale_reasoning", "governance"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fec20c925e6b6663d6eef77d39076540095ee4ac"}
{"id": "train_24093", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "tooling", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7e379298ab2db9951d04cd57181cb87c8316703c"}
{"id": "train_24094", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["auditability", "documentation", "tests_are_truth", "governance"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "427dd56958ef7487f132f5eff3b944f9a15168fb"}
{"id": "train_24095", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "security_gates", "tooling", "documentation"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cd7d7d8a81fe2e056438dc183a916cecce22feeb"}
{"id": "train_24096", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "ci_integration", "evaluation_metrics", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8999a79cfbb6edb6ac392ff2c1f29f36ebdb855b"}
{"id": "train_24097", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "security_gates", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "62958336c6704bcfe8285ee6fb25d6036cbf64d5"}
{"id": "train_24098", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "ci_integration", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fe40b4e5f2e268ace1aa388f0ec6bf331fdc6c25"}
{"id": "train_24099", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "reproducibility", "ci_integration", "tests_are_truth"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d008a2e469279df12a991eca8e30a8b6c28d40f1"}
{"id": "train_24100", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["auditability", "reproducibility", "evaluation_metrics", "tooling"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6ff9c40b027fba141b02f92bcdb88ca88d9247a8"}
{"id": "train_24101", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "tooling", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "adcd674dd2f8e8146c64a814931df57129109e67"}
{"id": "train_24102", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "governance", "cost_latency_tradeoffs", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "130dbe21f55b7af19c6763a38173bddc70d7c7ca"}
{"id": "train_24103", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["auditability", "security_gates", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c46485a8df00a56ec69114eeceb79c25a1828d0d"}
{"id": "train_24104", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "reproducibility", "documentation", "tests_are_truth"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "61337214327c3d2caf5f228b0a487dc508f4c76b"}
{"id": "train_24105", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "reproducibility", "security_gates", "auditability"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "692a2f379fbd810d22d5093d55fc18c012c01815"}
{"id": "train_24106", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "security_gates", "governance", "documentation"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "972a84295928e8211afa3c396072ae6afa9859c0"}
{"id": "train_24107", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "reproducibility", "tooling", "governance"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fe12cc3492f49edb2ddc23dcbbcbb423c11d6230"}
{"id": "train_24108", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "governance", "tests_are_truth"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "90425b9bb2887d3df59e98d345d3fa06bd3599ac"}
{"id": "train_24109", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "tests_are_truth", "ci_integration", "tooling"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "eea75b7a021763c1479b7229aaaf017981977d48"}
{"id": "train_24110", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "reproducibility", "auditability", "documentation"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f45ebd61862149097c76b2b6410999b4a17c11ad"}
{"id": "train_24111", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "evaluation_metrics", "tooling", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0094d78ead2b3fcec7fd30b15f95b789f199b466"}
{"id": "train_24112", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "tooling", "cost_latency_tradeoffs", "governance"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1f2fa10271f46b3ec82eedc94be90da5bd2dd07b"}
{"id": "train_24113", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "reproducibility", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ba72e47d5f99ba421dc4a34547c9351f69e873ac"}
{"id": "train_24114", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "evaluation_metrics", "tests_are_truth", "governance"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a0499e1ef1f8becad6efdf3d574c1aa0ef73b380"}
{"id": "train_24115", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "cost_latency_tradeoffs", "security_gates"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e60a641ca130837a8025326e7acafa36bfe2bb4b"}
{"id": "train_24116", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "governance", "repo_scale_reasoning", "tooling"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "27ec5bceaac023c5736839ad753e79055a04c8d5"}
{"id": "train_24117", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "tests_are_truth", "documentation", "tooling"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9d733940fd5f64b8f073ddd4120a6b0bf2eab677"}
{"id": "train_24118", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "ci_integration", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8013d602e6bdb6f1f5669ebab131bcecbed4e7f7"}
{"id": "train_24119", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "repo_scale_reasoning", "tooling"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c678188630228e2a0096570432ff8a0519012ec4"}
{"id": "train_24120", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "ci_integration", "repo_scale_reasoning", "auditability"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2c65b8ee78d53c7a2936cb21b4f61b4df17951ea"}
{"id": "train_24121", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "documentation", "tooling", "reproducibility"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d1a9ce8902f2d557ce6b33bc3069eafeaac9eab1"}
{"id": "train_24122", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "documentation", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ea257bba8879347bf6a81703cb30b1dfc9a7129c"}
{"id": "train_24123", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "reproducibility", "tests_are_truth", "auditability"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "627a2f76f9d58fe70051a00adbdf2c6eaacdc1b8"}
{"id": "train_24124", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "repo_scale_reasoning", "security_gates"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "62e68d8da4a2c5627784ff1e6898a9174697a401"}
{"id": "train_24125", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "documentation", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6b7d468dfe32ebe409063c2126423a0d4589f88a"}
{"id": "train_24126", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "tooling", "governance"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7fb53b304994a1e7327b4d6c9d0b403e161190fc"}
{"id": "train_24127", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "reproducibility", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6757200439f8f444c86e534cb64b75ff62e0c358"}
{"id": "train_24128", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "reproducibility", "ci_integration", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6a6d663919b6e454837c0861a3bbf8f37cd1a935"}
{"id": "train_24129", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "ci_integration", "documentation", "tests_are_truth"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ea5e74a9e8dfdb14be0b309bde7b5e639711fb2b"}
{"id": "train_24130", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "governance", "security_gates", "documentation"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ae5578f81b9c61e1586714f781d695062df6a3ec"}
{"id": "train_24131", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "auditability", "governance", "documentation"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dabbb09bf2db2b2db24e8dd45ea6a6c39c35c4d0"}
{"id": "train_24132", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "tooling", "auditability", "ci_integration"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7bf282ab27819eb72b91127540cf2a6114c29765"}
{"id": "train_24133", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "tests_are_truth", "auditability", "ci_integration"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f1058f5215024e94943ffea4f91d51f1af4b1233"}
{"id": "train_24134", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "cost_latency_tradeoffs", "ci_integration", "governance"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ca69bbebfe5cdf1a011f62eca392f4e3137b687d"}
{"id": "train_24135", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "tests_are_truth", "repo_scale_reasoning", "governance"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "eab6106ab0e88a5c848bd5cb1687102dd4e5dcde"}
{"id": "train_24136", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "governance", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "028587454e1e70fc03f16ebd05c31f831bede196"}
{"id": "train_24137", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "ci_integration", "evaluation_metrics", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bcd4534bfd0227f4f1f30ef370ef6d4798a00c52"}
{"id": "train_24138", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "evaluation_metrics", "governance"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "63465dcff70f5b810c491b426563577d3dbe0dd3"}
{"id": "train_24139", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "ci_integration", "reproducibility", "tests_are_truth"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9dc3aae177e802b1b44ac02cba8870945b15f12d"}
{"id": "train_24140", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "governance", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "450ff3e9c51be3e91020ade42baa233e78c84641"}
{"id": "train_24141", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "tests_are_truth", "ci_integration", "tooling"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "72bb94cc652f73579b879cf14f872ac0c01b4507"}
{"id": "train_24142", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["documentation", "governance", "ci_integration", "reproducibility"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8e84758ee125f7a3fb40781060b064a698dfb545"}
{"id": "train_24143", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "repo_scale_reasoning", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6d8ef1ee7e2035551735d6de8238aabf3b4c4fe3"}
{"id": "train_24144", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["governance", "auditability", "documentation", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "93e0d1441484afb7d1c67cc84c93dcebaa241545"}
{"id": "train_24145", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "reproducibility", "evaluation_metrics", "documentation"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "312cc410e48b870011618a85cbc3b7a229943cbd"}
{"id": "train_24146", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "auditability", "ci_integration", "evaluation_metrics"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f81cef50d0962ec1828ceed32be6585aafbf9433"}
{"id": "train_24147", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["governance", "tests_are_truth", "cost_latency_tradeoffs", "security_gates"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "52f7671716b3a2f2121da9558aefb15447e23598"}
{"id": "train_24148", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "cost_latency_tradeoffs", "governance", "tests_are_truth"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "11f0c377ff8d9aab111e60239df3d0163237d274"}
{"id": "train_24149", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "repo_scale_reasoning", "auditability", "governance"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c4970e35b3af102327ecbcf4f6fe4323cfd87405"}
{"id": "train_24150", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "tooling", "ci_integration", "reproducibility"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d2af80b7c7ecc6ef04facbb64f0c04e017175147"}
{"id": "train_24151", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "security_gates", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c74a23897e86583c0a127778f45dd501330dcdaf"}
{"id": "train_24152", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "documentation", "tests_are_truth", "auditability"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a6936664fa8483934989ae61f9efe413a188d138"}
{"id": "train_24153", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "tests_are_truth", "documentation", "tooling"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6b30728a03c77a55ee2c129d143beac08f9edcde"}
{"id": "train_24154", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "ci_integration", "evaluation_metrics", "documentation"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8da6080d73f1970c545a5968b4f6663c3b09c8df"}
{"id": "train_24155", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "documentation", "security_gates", "reproducibility"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7d13a413377240baaf064bbb972eaca22268c175"}
{"id": "train_24156", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "documentation", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2776a6cb8004d80832382aed646f6b733cc6c022"}
{"id": "train_24157", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "tests_are_truth", "cost_latency_tradeoffs", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "578cc0a47b2dc2bd9a77fa2ac868deb91c7d6c8a"}
{"id": "train_24158", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "documentation", "auditability", "evaluation_metrics"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "acf6fc05c7ad5c96f1f5c0effcb62953ba9a5734"}
{"id": "train_24159", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "governance", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "be10c36d5db9b6ab448e22e136c4e1e186cc71eb"}
{"id": "train_24160", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "ci_integration", "evaluation_metrics", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3f8eaa5883e2f77e0aab8df5f9399103b6ba8f64"}
{"id": "train_24161", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["governance", "auditability", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fd7b69a1104e0744cfb116e855909a64cc1a91d9"}
{"id": "train_24162", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "reproducibility", "governance", "tooling"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bfd2ecbbbca042eea311e87b4cae207e60ebfe80"}
{"id": "train_24163", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "evaluation_metrics", "governance"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9648fe749ce74968945762ac7559a00b68e1bf73"}
{"id": "train_24164", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "governance", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e22ed2fe61bf20f1ceae1e06ce1f91daad6fa353"}
{"id": "train_24165", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "security_gates", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "84603585ce645f45cafc952f9bacaaa31c2e7d4e"}
{"id": "train_24166", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "cost_latency_tradeoffs", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9e587a47e73b001f05ce29bded2cf1b210f4d1b6"}
{"id": "train_24167", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "tooling", "tests_are_truth", "governance"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ad363dc07fb084cec9429b75ce0c702286a0c7d5"}
{"id": "train_24168", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "governance", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dac52d4eea594e8a9395f237046d1262fa0c2f09"}
{"id": "train_24169", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "tooling", "reproducibility"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "60fd695dcd44e717bce66b5b1fbb61540525c9f1"}
{"id": "train_24170", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "documentation", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b4ff8b8a97f8fcd08529bfd5474f810b7b33d3f3"}
{"id": "train_24171", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["auditability", "governance", "security_gates", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5830fb77073abb14424b558e25060550e9907934"}
{"id": "train_24172", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "security_gates", "ci_integration", "tests_are_truth"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7823147a4a63257df8da6a46aa028bc1132272b7"}
{"id": "train_24173", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "documentation", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "73a2e8dc6478042d7f5c43601999d461462ff75d"}
{"id": "train_24174", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "ci_integration", "documentation", "auditability"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cc1f6dce182bdbda2994192940f591b0ed5b529b"}
{"id": "train_24175", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "security_gates", "cost_latency_tradeoffs", "governance"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9ad0a2d9c815ca22715e12e8bf5003cdb696a9d1"}
{"id": "train_24176", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "auditability", "governance", "evaluation_metrics"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e63d09a7d58e760397b058ded7a490052edeff02"}
{"id": "train_24177", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "cost_latency_tradeoffs", "documentation", "auditability"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6a712f43dd1a65061cddd123a2865769f40b8346"}
{"id": "train_24178", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "governance", "documentation", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "78029f7c270a35c2a7c7666690577cbf7972191d"}
{"id": "train_24179", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "auditability", "ci_integration"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "233bc64b0f9f1fab7a54d7b8561ac5d20a833cf6"}
{"id": "train_24180", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "governance", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a6a1a95d75e0317bf4577cc566d2e9cf0b82cceb"}
{"id": "train_24181", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "ci_integration", "evaluation_metrics", "documentation"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7bea60fd5d007e583b43400c4afbdb27f4ff8d86"}
{"id": "train_24182", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "evaluation_metrics", "security_gates"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "90e67067a2e9cc4c1f41b37ef01c0dc2d8d96d51"}
{"id": "train_24183", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "auditability", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "eb20e2f0e1bbbd956b4233ecdd18e379120224c0"}
{"id": "train_24184", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "repo_scale_reasoning", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "aabf985f25b63adcf4442f1386d13663ef03fc6b"}
{"id": "train_24185", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "tooling", "evaluation_metrics", "security_gates"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6197dd75b7938446ade53d5846aba8f1b28c74fe"}
{"id": "train_24186", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "cost_latency_tradeoffs", "documentation", "ci_integration"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b1283313eef4a13da415618c733be68f3439799f"}
{"id": "train_24187", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "governance", "tests_are_truth", "security_gates"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bab7cd3b1aacb4de1e9a5e47155856b54291671f"}
{"id": "train_24188", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "auditability", "documentation", "ci_integration"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fe1db242540fd4ecb0494502653be21e718064ef"}
{"id": "train_24189", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "security_gates", "ci_integration", "documentation"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "191f89d51ddafa2fc97d018445ba00133b48aca0"}
{"id": "train_24190", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "governance", "reproducibility", "tests_are_truth"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3774de09a4482467a43ee00f34f990ec261f3536"}
{"id": "train_24191", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "tooling", "ci_integration", "documentation"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "877ce1a5a3dba96fc1ce4b602e5008fbf061bc46"}
{"id": "train_24192", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "auditability", "cost_latency_tradeoffs", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "763e1069722b95625d79f56fe0d0bfb35e3a12bc"}
{"id": "train_24193", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "evaluation_metrics", "auditability", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "69ec6251d327eac16e427befb43542a50d30ef4e"}
{"id": "train_24194", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "evaluation_metrics", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9567105427f993d78e18f7ed95a038cb021efa47"}
{"id": "train_24195", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "documentation", "cost_latency_tradeoffs", "repo_scale_reasoning"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4904c4351679a8dc21a0ebd86bf0c59441b216a1"}
{"id": "train_24196", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "reproducibility", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7fec6b2f6a8ee23db8d7a097de2e64aed1885a6d"}
{"id": "train_24197", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "documentation", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ffaa975249201c50e381616ac4db705cc1e84fdd"}
{"id": "train_24198", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "security_gates", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7583497c97360104f2198f228833cc36e221dac2"}
{"id": "train_24199", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "governance", "ci_integration", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "28b8c7a8d41505f8d36c04323ecdda1b8d6d7c57"}
{"id": "train_24200", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "repo_scale_reasoning", "evaluation_metrics", "governance"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9687604bafe64fa885857b0e8253766b43d76b42"}
{"id": "train_24201", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "repo_scale_reasoning", "auditability", "security_gates"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ea2eea4f6e76b4ec90327ebc4abc37fc851f91bd"}
{"id": "train_24202", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "repo_scale_reasoning", "governance"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8538f903e427d1dee014983054cfaf2ee440c031"}
{"id": "train_24203", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "repo_scale_reasoning", "documentation", "tooling"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cbb08e283bdf529dbfab297be21710434737f52f"}
{"id": "train_24204", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "tooling", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2fe67926d0a0abeceb5bb24926a19eeaab6d1c12"}
{"id": "train_24205", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "tests_are_truth", "security_gates"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9c64611c81f77f9277ef7a93fafbff2fd86c23ce"}
{"id": "train_24206", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "repo_scale_reasoning", "documentation", "tooling"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f1af06da8ccdb0c7aa2f0a8dca6b81b263a8fdde"}
{"id": "train_24207", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "governance", "tooling", "evaluation_metrics"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1f30ff39b632f4c4a6953a50c07ebb25d7eaa108"}
{"id": "train_24208", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "auditability", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b635f408caa4c8da3be8f02cde75e3faa5edc6db"}
{"id": "train_24209", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "security_gates", "governance", "tests_are_truth"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a5681f585f0c7f643ede68d01e2fd4397659e6c4"}
{"id": "train_24210", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "auditability", "security_gates", "ci_integration"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "982ba2fa2e70020438f68e3b10ab27cc96cc9966"}
{"id": "train_24211", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "security_gates", "ci_integration"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "77f433e416f6d40c16f24469f80ae7c8f7b0794b"}
{"id": "train_24212", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "auditability", "security_gates", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2460d23d52eea0d96bb62890af25a1989738858f"}
{"id": "train_24213", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "auditability", "reproducibility", "tests_are_truth"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5b0fd8a4558de4f90a96e5c2070124656f970910"}
{"id": "train_24214", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "security_gates", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "97a3a899314ae5561eb57a5345fe5253488f20a3"}
{"id": "train_24215", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "documentation", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5f9994cec5746f47161cb696652da0d045485ebe"}
{"id": "train_24216", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "cost_latency_tradeoffs", "tooling", "documentation"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ae8c3874f1ee653a3d5475933dc1979bef5bd254"}
{"id": "train_24217", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "auditability", "tests_are_truth"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "98fc2fef1e6075c48b7e6a0708797dbbc668ab0c"}
{"id": "train_24218", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "tooling", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ada731b403ea98a924db7a1556916952f223e894"}
{"id": "train_24219", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "documentation", "auditability"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e88c2b81d210e395eedb1992eb42cf232ec0c6df"}
{"id": "train_24220", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "tooling", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8747185a7f5d61f21905c1e239542135ba6ed9b2"}
{"id": "train_24221", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "ci_integration", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "96fe001e36bc7613f8909ada150d22c0aa6a3306"}
{"id": "train_24222", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "governance", "reproducibility", "tests_are_truth"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b4f06195267de0e693aa1b37df2f4c8f771b0c4e"}
{"id": "train_24223", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "reproducibility", "security_gates", "tests_are_truth"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9a9b236a7f47ce78277b375f8643fc5add5956f4"}
{"id": "train_24224", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "cost_latency_tradeoffs", "ci_integration", "governance"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9a9e156bc5c95c12ea263b4c50c91744e551e2e6"}
{"id": "train_24225", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "governance", "repo_scale_reasoning", "tooling"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a6936664fa8483934989ae61f9efe413a188d138"}
{"id": "train_24226", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "documentation", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "23a1a42b97b53afad135f5e65529298b0bc9c9fd"}
{"id": "train_24227", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "governance", "ci_integration", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6ad15cc3eeb3cda2c220530a8634429eaa0a2d13"}
{"id": "train_24228", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "evaluation_metrics", "repo_scale_reasoning", "governance"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8feb00e71854b4e5163b6d07ed523d908c92a4b9"}
{"id": "train_24229", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "tests_are_truth", "reproducibility", "evaluation_metrics"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cb04447c2ec5ff2c9a5861cab4f72b9757554061"}
{"id": "train_24230", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "ci_integration", "evaluation_metrics", "security_gates"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "26d8627909420055e975fd8dc04ec0200cb2f85f"}
{"id": "train_24231", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["governance", "security_gates", "tests_are_truth", "auditability"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3a4fb9fe20d507d8adf3fdcbbf854afc2329ec08"}
{"id": "train_24232", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "security_gates", "reproducibility", "tooling"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f59a78cdda57663dd94d08be69bf0907ab7b4115"}
{"id": "train_24233", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "tests_are_truth", "evaluation_metrics", "ci_integration"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c63cba48739fbac651ba47f59dad4536680b3f55"}
{"id": "train_24234", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "governance", "tooling", "tests_are_truth"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "82671206827793e1fd765d879f63e1eeb24a2dfc"}
{"id": "train_24235", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "documentation", "tooling", "evaluation_metrics"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d44110c63f93c3387d6af4b66d9237e75771f61e"}
{"id": "train_24236", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "evaluation_metrics", "tests_are_truth", "security_gates"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "41bb02fc88a6b3d3aa0a98b642bdf787b3f43b92"}
{"id": "train_24237", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["auditability", "tests_are_truth", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4bf30987a0da1d737b1a306ae57a510e3303c3d6"}
{"id": "train_24238", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "security_gates", "documentation", "ci_integration"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0e33d3ae7d3dfb6c281dbf012bf0f4543eeab25d"}
{"id": "train_24239", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "documentation", "governance", "evaluation_metrics"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f8adc6d212148ad77e4779053a5821dad6c77c9c"}
{"id": "train_24240", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "reproducibility", "evaluation_metrics"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b1460b623022c7ff9f6a4c3794f25bddd3dacf43"}
{"id": "train_24241", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "repo_scale_reasoning", "tooling", "security_gates"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cf99a3f20beb7b56eb3dce6ebd4848db7458f0ec"}
{"id": "train_24242", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "governance", "tooling", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e21b7b053e6f5cebb67a6d39c18d2dc856b6a8d6"}
{"id": "train_24243", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "documentation", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "205b10828e58b0c7b28dca5397e808f39b13149e"}
{"id": "train_24244", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["auditability", "ci_integration", "reproducibility", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3ea22a60edd999adff1a0e9d1618a9eb3e18e517"}
{"id": "train_24245", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "governance", "tooling", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f6257e135e3a935e125b21c191eff4e0007dcfe0"}
{"id": "train_24246", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "tooling", "governance", "documentation"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bcbb7373a2a6bcf36ddfe19ff331aed0c7162bdc"}
{"id": "train_24247", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "tooling", "evaluation_metrics", "auditability"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "05949bdfe468fdbdd8a837e68d5d7e9f0ded2f46"}
{"id": "train_24248", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["governance", "evaluation_metrics", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "016f2e27d541a4b56ed5171fe9fbbc211bd42e98"}
{"id": "train_24249", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "auditability", "documentation"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "75c44cb56f0a07e5cd716d972fb6968c01514c3b"}
{"id": "train_24250", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "repo_scale_reasoning", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ab4d4e9983bdd980f4b56b6e2785f9beb8a772c5"}
{"id": "train_24251", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "evaluation_metrics", "security_gates", "tests_are_truth"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cd537e71f05b385800d47cce9dae25c82a2910d9"}
{"id": "train_24252", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "tests_are_truth", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "530e1480e1a55cf0be11672b081ca4491c811a7e"}
{"id": "train_24253", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "cost_latency_tradeoffs", "auditability", "reproducibility"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "17d2e9262759c5ccb65082fca2317f221296dadc"}
{"id": "train_24254", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "tooling", "auditability", "governance"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "672e9f4a5f927846dc1669585d9676175bc8aebf"}
{"id": "train_24255", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "governance", "reproducibility", "ci_integration"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b9c742c2348a747d74d9ac588052ee8645a284a3"}
{"id": "train_24256", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "tooling", "auditability", "documentation"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "55a869b6853401241f18c95dc0fe0703306da9d5"}
{"id": "train_24257", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "auditability", "documentation", "tooling"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6fd6dba57b7e24fa8a907383c8ff51b7255f3ecd"}
{"id": "train_24258", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "reproducibility", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ea7e3f5169a4f88ab89f7ab839bb7f6005718e53"}
{"id": "train_24259", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "tests_are_truth", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "86984e5570e53bfeaab5448927e980f9fc1ee783"}
{"id": "train_24260", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "governance", "documentation", "ci_integration"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1e86d8dffd0b36132b4b7a7e2ac479c3c4438d95"}
{"id": "train_24261", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "tests_are_truth", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1c110d36471bfbb79dbfd23490e14b042c4dd063"}
{"id": "train_24262", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "tooling", "reproducibility"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9a4fe59eac2d6572b993e72f8cd96eb94d7c19d0"}
{"id": "train_24263", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "auditability", "documentation", "security_gates"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b69040b81f77c463b6eed544558f6c805fc04e39"}
{"id": "train_24264", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "tests_are_truth", "tooling"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e715aa4bd732dcaabf2807383f53f719661e3593"}
{"id": "train_24265", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "reproducibility", "ci_integration", "security_gates"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "485506c03e0bef3767d260f20a47b90125f32024"}
{"id": "train_24266", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "repo_scale_reasoning", "evaluation_metrics", "governance"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f0b713c6cd9de7891b27449a08bc552013ca71de"}
{"id": "train_24267", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "repo_scale_reasoning", "tests_are_truth", "documentation"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "87d085d6f4e81e754c94ba6e233b005a32e7f1f3"}
{"id": "train_24268", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "ci_integration", "security_gates", "auditability"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8a5423890e3be871a44bca7138b3b4b5302f82b4"}
{"id": "train_24269", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "reproducibility", "security_gates"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6abb2a0a3be4e5eaa1a0bae1cdee4f91798dc6f4"}
{"id": "train_24270", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "governance", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "480aa49732080e13b97bb705aae6f27fa66443e3"}
{"id": "train_24271", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "auditability", "evaluation_metrics"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bfac78da49ea0d091a4beaf309bf2866ef8fe087"}
{"id": "train_24272", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "ci_integration", "tests_are_truth", "security_gates"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "788b7e2ce80014ea8c7816fd4b563a6399d01004"}
{"id": "train_24273", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "ci_integration", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "84bf7e4a47f27d27ec7c6f303266e921d116d1dd"}
{"id": "train_24274", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "tests_are_truth", "repo_scale_reasoning", "auditability"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0cbda5b5948afb982797e8143043b1f34d086d61"}
{"id": "train_24275", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "reproducibility", "tooling", "governance"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "62582131a22a8bbcdbcc0849769d3a6847281ae4"}
{"id": "train_24276", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "documentation", "ci_integration", "governance"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9d75dd0fae0e63704a3807d112db21581443ebed"}
{"id": "train_24277", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "auditability", "tests_are_truth", "documentation"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c4474e925f89f10b086c7e0f4fe4382f709a33da"}
{"id": "train_24278", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "tooling", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8ba751efd2cc7ef6c3ee460e437bf13b6b5f8656"}
{"id": "train_24279", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "repo_scale_reasoning", "evaluation_metrics"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9d17620bd6b26264d245333eac2c78541f64c400"}
{"id": "train_24280", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "security_gates", "tests_are_truth", "auditability"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "99460f5977d95444374310972cf25a20c8b496fc"}
{"id": "train_24281", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "repo_scale_reasoning", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "83ccceb5222ad0e7e51b6973a26ce8815113a22a"}
{"id": "train_24282", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "governance", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1a8b59dbdb0f3eb3b45859e3a7fd90c11d1efc08"}
{"id": "train_24283", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "governance", "ci_integration", "documentation"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5dd87bfaed1056bed87bb53b9556f74ef04ba67a"}
{"id": "train_24284", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "governance", "ci_integration", "evaluation_metrics"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "90c4171bf325ba0030a0a7c6003bf1aaea8c00c5"}
{"id": "train_24285", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "tests_are_truth", "repo_scale_reasoning", "reproducibility"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "43b246963c9f860c86c8c618af834b504b172725"}
{"id": "train_24286", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "evaluation_metrics", "documentation", "tests_are_truth"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "72b9a148aeb944987bef821ad5644c65de7f2209"}
{"id": "train_24287", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "reproducibility", "documentation", "tooling"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "54031425e0e64b261b2d535a77ab6bd15b6a7473"}
{"id": "train_24288", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "repo_scale_reasoning", "tooling", "ci_integration"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bedae43e68053d57dd9e31bb15eab562e7d41b0d"}
{"id": "train_24289", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "auditability", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "537d07f675d3dbd5f9732796952fb25dc70f2eb1"}
{"id": "train_24290", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "evaluation_metrics", "tests_are_truth", "tooling"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b1274ef6a56b6be5dcc2b83c1a463a58364b26f2"}
{"id": "train_24291", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "tooling", "security_gates"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c130c33cadaf3fb138d92e26b47042a19294b626"}
{"id": "train_24292", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "security_gates", "documentation", "auditability"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5a0a0215a163b0be574858bcdfd0dcbb0ac13459"}
{"id": "train_24293", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "reproducibility", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f9c9956e071a1b4e31272ea0d096924895745bd7"}
{"id": "train_24294", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "repo_scale_reasoning", "security_gates"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "872effa4de644390e02659809f67d30004f17ce2"}
{"id": "train_24295", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "tooling", "auditability"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8fe15aae8c00e7055050a921e9b07acdb59c2520"}
{"id": "train_24296", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "governance", "reproducibility"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "658f171cbb5f93e4b993cb52c63c96b915af861d"}
{"id": "train_24297", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "evaluation_metrics", "auditability", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ca8145fd6cb788e00b9de2ba7961db651aacb310"}
{"id": "train_24298", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "cost_latency_tradeoffs", "security_gates", "reproducibility"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e581359e4bd951f6010a9e53e9201e113708e47c"}
{"id": "train_24299", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "ci_integration", "security_gates", "reproducibility"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ab8909e8368bce74bd7fbbcc48f8bb266ea206c5"}
{"id": "train_24300", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "governance", "reproducibility", "documentation"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9665adb8e23ed06d831b6e5add4b1dd253400cef"}
{"id": "train_24301", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "tooling", "reproducibility", "security_gates"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d712b9d0015428720c8f5b4f86f72131ed4c9d9f"}
{"id": "train_24302", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "ci_integration", "auditability", "tooling"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "eee0353682455c824cab04b8e420414c84113b13"}
{"id": "train_24303", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "reproducibility", "governance"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "13376c2de37ebd59bf1cc4ceb61a3b820cf4c586"}
{"id": "train_24304", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "evaluation_metrics", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1a8b17b1cb74705fb2940a0c8fb16a70a73d2581"}
{"id": "train_24305", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "evaluation_metrics", "documentation", "tooling"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9124e08a0b3d748bd1de620c45ac4bf5970e98bd"}
{"id": "train_24306", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "governance", "tooling", "repo_scale_reasoning"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f856e4bc5a75a2c32a7dd000ef986a08e6ddacaa"}
{"id": "train_24307", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6552e60735050a2db9005e4337220a5fc05cd814"}
{"id": "train_24308", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "tooling", "repo_scale_reasoning", "reproducibility"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dddcfcdb9afc4fce0f99be37361a95472d5f565c"}
{"id": "train_24309", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "tooling", "reproducibility", "tests_are_truth"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c53c8266a7b0582add953313b10cb3c193109e6f"}
{"id": "train_24310", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "governance", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "62f0bcad36891d67012cea4a580bb2f157419ca0"}
{"id": "train_24311", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "security_gates", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "de90b6c96ff9bf99c400c8287b3d9da417c31315"}
{"id": "train_24312", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "documentation", "cost_latency_tradeoffs", "security_gates"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "191e1fe0681a5a372407c3bb965b98386872c2ee"}
{"id": "train_24313", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "governance", "reproducibility", "tooling"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a447e31eafdceae7c2703de170665e37b25b0328"}
{"id": "train_24314", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "tooling", "governance", "repo_scale_reasoning"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "32650b9256df9ea914e42d7130e2ccbdc91486bd"}
{"id": "train_24315", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "cost_latency_tradeoffs", "security_gates", "reproducibility"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5e0ac3c847ef0bd833b359c7354f5b00df5797ec"}
{"id": "train_24316", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "governance", "tests_are_truth", "reproducibility"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "95b1435a52b0cc9e67de0a4f6ad5f03e2bd76539"}
{"id": "train_24317", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "tests_are_truth", "ci_integration", "auditability"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f1055eba9ad97437efd936a4212791b98883d0a4"}
{"id": "train_24318", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "documentation", "security_gates"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4a01dac3255fe0f380672842133a95f7beed8fdd"}
{"id": "train_24319", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "cost_latency_tradeoffs", "ci_integration", "documentation"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "778cba46161e590bcc0120b57a488f4229fb62d6"}
{"id": "train_24320", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "governance", "reproducibility", "documentation"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "49ce12f46d7407c0d9f6a15ec0cb3f784be09ca8"}
{"id": "train_24321", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "ci_integration", "governance", "tests_are_truth"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b246eed34cd714360fff5b7da9be7ffed5dc6735"}
{"id": "train_24322", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "tooling", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7502cbb4479b20fcc829b5d567d238f481c81c43"}
{"id": "train_24323", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "tooling", "governance", "security_gates"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1a41ae08c43113256364e498e33daff6da212734"}
{"id": "train_24324", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "ci_integration", "cost_latency_tradeoffs", "repo_scale_reasoning"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b56d0f41b686ff142ed422b0316ef56c4a9440c4"}
{"id": "train_24325", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "governance", "evaluation_metrics", "documentation"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5ec3709e67816cc979eb6f5e5e56e07068a32b84"}
{"id": "train_24326", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "auditability", "ci_integration", "tooling"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "772014e51e61279602e4156527fe9ce029b23b39"}
{"id": "train_24327", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "reproducibility", "auditability", "tooling"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ecafb6cd0f25506751884792e634b5fe977aabd6"}
{"id": "train_24328", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "tooling", "governance", "documentation"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a14d7ea6b226f3b8a18df7227ec1b2cdd7d842e5"}
{"id": "train_24329", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "tooling", "reproducibility", "security_gates"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2f757d7b535e3a74d55d9957ed2528680966bbce"}
{"id": "train_24330", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "governance", "reproducibility", "evaluation_metrics"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5038130dfeb6bb9b5dfbf4a3c2681a16f2944186"}
{"id": "train_24331", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "tooling", "documentation"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "79fd8daab2dd4d02dc682cedad5e6e1e68a389ad"}
{"id": "train_24332", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "ci_integration", "security_gates", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "39ceb873ddcd250b3df54f7af3be934bff6cf816"}
{"id": "train_24333", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "reproducibility", "ci_integration", "security_gates"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1c9e1515bd2c1e2596491460aca7083b66d20b18"}
{"id": "train_24334", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "tests_are_truth", "governance", "reproducibility"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2461c9ef5a9367775fe09c94ba73b2ac176c460e"}
{"id": "train_24335", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "tooling", "security_gates", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cc1305e1ebbca9bd7010bd17654bbda6e03106e9"}
{"id": "train_24336", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "auditability", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3125d830fc2079ea4a3ca2e565e863bfbd2d14f9"}
{"id": "train_24337", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "security_gates", "auditability", "tests_are_truth"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d0d7843dfee06a611b425770d0b8eb175e15f63f"}
{"id": "train_24338", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "auditability", "tooling", "governance"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f2f4ea26f5f2b24dbaf4f97a5839ea9f8ead2b76"}
{"id": "train_24339", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "repo_scale_reasoning", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "028c30d4fda3a86dc74016af63e9ecde2b169a5b"}
{"id": "train_24340", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "ci_integration", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d7dea45a3956514e131af7f551082ca09d1cccac"}
{"id": "train_24341", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["auditability", "security_gates", "ci_integration", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "473ddc210ecad46b6e778f25db852b61227ca347"}
{"id": "train_24342", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "tooling", "evaluation_metrics", "documentation"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0bb9efa042c3bb4b5af881944f4535ae340cde24"}
{"id": "train_24343", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "tests_are_truth", "auditability"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "49a0140da7f00da31d3fd34bc12389dd311bcc6f"}
{"id": "train_24344", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "governance", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c65f7cbbf32747ec3f3c7cba4fcfc58172f677df"}
{"id": "train_24345", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "tests_are_truth", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "14b0b5f3da3d0f388ef8efe7078c7209fc06d1ee"}
{"id": "train_24346", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["auditability", "reproducibility", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f27bc74bec535b1318c77c05bde5e757cef0cdd2"}
{"id": "train_24347", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "documentation", "auditability", "governance"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "06d22ac1448f4499d0cdf6bd91f55df5bad2cc51"}
{"id": "train_24348", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "auditability", "repo_scale_reasoning", "reproducibility"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6b464b89e9944f0f157397f195eda20614f07e62"}
{"id": "train_24349", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "repo_scale_reasoning", "documentation", "governance"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a2d9d97ba66139ac53d7040343408a0e9992efcc"}
{"id": "train_24350", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "ci_integration", "tests_are_truth", "governance"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "078ac111116620df8be2942f8ccaf7eecde43577"}
{"id": "train_24351", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["governance", "repo_scale_reasoning", "evaluation_metrics", "ci_integration"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6eddf3a1b7ca8eb6b9050978633cd4f2c4da4f7d"}
{"id": "train_24352", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "documentation", "reproducibility", "governance"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5181c0cf1f0c52c1fffc1af98f389ab308fe0bd7"}
{"id": "train_24353", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "repo_scale_reasoning", "tooling", "auditability"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f3a715e3150c1709eaaa20d73cd825a468e8f9d7"}
{"id": "train_24354", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "documentation", "ci_integration", "security_gates"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bd6300f4313b4713c4e975a56133b70ef13e1a40"}
{"id": "train_24355", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "repo_scale_reasoning", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3e546e8f50b831d71b6321a895435702452dcaf4"}
{"id": "train_24356", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "auditability", "governance", "documentation"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "70c053d3dd319c7b830e88d14dc42322c27eedb2"}
{"id": "train_24357", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "documentation", "tests_are_truth", "reproducibility"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0595739edf2316af653519ca1276c21e6fbe762b"}
{"id": "train_24358", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "reproducibility", "security_gates", "governance"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a3e75d2fe3be9383c583da3f8374e27c96568c27"}
{"id": "train_24359", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "ci_integration", "evaluation_metrics", "documentation"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4a46d231c653209b604edeff9ae9e2c3ad16fac4"}
{"id": "train_24360", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "auditability", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "56d19a45342af4a11337112355dde51c0805b628"}
{"id": "train_24361", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "cost_latency_tradeoffs", "ci_integration", "tooling"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b9e7caa9338433b2df0aafa207167c1fc547268f"}
{"id": "train_24362", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "documentation", "auditability", "tooling"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d7549027b2efc438f0773e9bf00c2904cf7e6d7e"}
{"id": "train_24363", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["governance", "auditability", "reproducibility", "evaluation_metrics"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d52f97f544c0e78cb41fd80b011e07fcbbefd951"}
{"id": "train_24364", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "reproducibility", "auditability", "governance"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5bd04ffe32ee45feac7a8dcbce24d6b4d4a65aa2"}
{"id": "train_24365", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "tooling", "ci_integration", "auditability"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "927333f72c74d5389ab6d35ca85a21beaaecfa52"}
{"id": "train_24366", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["documentation", "tests_are_truth", "governance", "auditability"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c313040843d4f031b5adcfd3ebdb606118d8cf6e"}
{"id": "train_24367", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "ci_integration", "cost_latency_tradeoffs", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "06dc65d1e1cacccc00f52533b9bb6937c53ffef9"}
{"id": "train_24368", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "ci_integration", "cost_latency_tradeoffs", "governance"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2e494702397a5b3200556ad7abfd14d439dcb960"}
{"id": "train_24369", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "tests_are_truth", "security_gates", "evaluation_metrics"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "54d3ddf0185bb6bdb40c166165a4aa58f5ae2a7e"}
{"id": "train_24370", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "governance", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "96628122349554184ddbf34934e53ecb61f791c0"}
{"id": "train_24371", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "cost_latency_tradeoffs", "tests_are_truth", "auditability"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cb5f62847f58a8c817adba67f4010d2a6fb766b5"}
{"id": "train_24372", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "documentation", "tests_are_truth", "reproducibility"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cf31a00501e73e7f74c47028c42c4e06689b540b"}
{"id": "train_24373", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "documentation", "cost_latency_tradeoffs", "security_gates"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "58a217617d67e72739d127da95dad95666cc88bb"}
{"id": "train_24374", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "security_gates", "cost_latency_tradeoffs", "governance"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "aa4b7db153849fc2b9d3b3959e76d25f2cbbac5f"}
{"id": "train_24375", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "tooling", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "93be6c6fd5a2c62c1d43cfcc7c249cd9fa26edbb"}
{"id": "train_24376", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["governance", "security_gates", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "79f4e3a828c03d4b9af03dde54a3ee87ffd606f1"}
{"id": "train_24377", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "security_gates", "ci_integration", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bc308d43addc17a93b804f0ec1fb3c26ec23450b"}
{"id": "train_24378", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "tooling", "governance"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bbe2d3c0b4fd59ad0c18190af9e2b81ae74751fe"}
{"id": "train_24379", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "tests_are_truth", "reproducibility", "auditability"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9a4caea74258787176aa9f6373d993e94b5fc329"}
{"id": "train_24380", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "repo_scale_reasoning", "evaluation_metrics", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0f2a1f6f8a82536ea8b31b2b248d34b5f8bafbda"}
{"id": "train_24381", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "tests_are_truth", "ci_integration", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a02b2f42c3bf8399da55be0189446fbffbc77216"}
{"id": "train_24382", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "governance", "tests_are_truth", "tooling"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "884de04cbe6cd9adff950e0cd38afcd5cfa24157"}
{"id": "train_24383", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "tooling", "evaluation_metrics"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1c0c60e3a8a285776598998b3a3202cff1f45249"}
{"id": "train_24384", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "governance", "reproducibility"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2f29a90ca7ca3241c29d23ac00fb800a8707a9ac"}
{"id": "train_24385", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "evaluation_metrics", "reproducibility", "ci_integration"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a7f536c8d99c8891b546a248bfe540e96af8a8e4"}
{"id": "train_24386", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "documentation", "security_gates", "governance"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "67bb1602766f14fc7668491da1ca5bafd22905f0"}
{"id": "train_24387", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "security_gates", "ci_integration", "repo_scale_reasoning"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "10db25b1e23163c93816e33359449476e7a728c7"}
{"id": "train_24388", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "auditability", "reproducibility", "security_gates"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6a30ea6b38b2f9a43da183f1a4c8af8cc6fd5ab7"}
{"id": "train_24389", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "tooling", "auditability"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "deaa760f0c630dd40d5b53a955db82d1b6f27139"}
{"id": "train_24390", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "governance", "reproducibility", "auditability"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e6dc38957091d61ddc36498791f7598f305a0676"}
{"id": "train_24391", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "reproducibility", "security_gates", "documentation"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dc084823dfb1aa2b946dd1595aacd98db83244fb"}
{"id": "train_24392", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "documentation", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "38dd9970493480e9438f95a0c842a2ec48ee6678"}
{"id": "train_24393", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "tooling", "cost_latency_tradeoffs", "governance"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "900ae3b290463430ea44772661f27cab53791566"}
{"id": "train_24394", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "reproducibility", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e690abdfb8e8e7506207eb3f6abdbc8018fb00d0"}
{"id": "train_24395", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "evaluation_metrics", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b568695557495741a4a6aa5dda9d3bbb4ff52e9a"}
{"id": "train_24396", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "cost_latency_tradeoffs", "evaluation_metrics", "tooling"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "092607ba435dd2dc05e1120472ed74749746001d"}
{"id": "train_24397", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "ci_integration", "auditability", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "75d62eaf8429f51178904a2e80f29f4b213a27c1"}
{"id": "train_24398", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "auditability", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "618e817e1851541678f726c55829a94f0cf04ddb"}
{"id": "train_24399", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "documentation", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0177f4ad086202d0303552e36666487bb7ea9723"}
{"id": "train_24400", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "tests_are_truth", "documentation"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "febb6c86f1d15ccd05990cacf9438e06006fdd44"}
{"id": "train_24401", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["governance", "tooling", "evaluation_metrics", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f4643c2174fe239473eebab5362cf2291c6af1d1"}
{"id": "train_24402", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "cost_latency_tradeoffs", "ci_integration", "documentation"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7b185884742440f41f0017609fef9962bfe8d62e"}
{"id": "train_24403", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "security_gates", "reproducibility", "ci_integration"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f7f9adee30364074f2631100f36d757930959c37"}
{"id": "train_24404", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "reproducibility", "ci_integration", "governance"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3b8ca3694f100ec11ad7ab964f74cc3b94fd30d8"}
{"id": "train_24405", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "security_gates", "documentation", "governance"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "151445fc598ab48337a5e0f202d07a0625c045fb"}
{"id": "train_24406", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "security_gates", "evaluation_metrics", "governance"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c0b13e1c0a04d90274646fd295946f173aa945f8"}
{"id": "train_24407", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "governance", "repo_scale_reasoning", "documentation"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8de0e11e91e4483ab3b1e2c5c92c30eb8d63e07b"}
{"id": "train_24408", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "tests_are_truth", "cost_latency_tradeoffs", "governance"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6561143872d901ff4ef3deb04d4d02b83bb64895"}
{"id": "train_24409", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "documentation", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "aab50aa8c1634ddcfc47aa69591c970b142b137c"}
{"id": "train_24410", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["governance", "tests_are_truth", "security_gates", "ci_integration"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "01144fd0b5aba05b7af9d58fa9cb55394493115c"}
{"id": "train_24411", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "reproducibility", "evaluation_metrics", "auditability"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a52d4defef28a8df8c1290ccfb4753290bed862d"}
{"id": "train_24412", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "security_gates", "tooling", "documentation"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "81f5b2fa7c2dae50067b301421ed27e38e6b6fcd"}
{"id": "train_24413", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "security_gates", "tests_are_truth", "auditability"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "329e879266246305cd654bff4f4919276163c1ba"}
{"id": "train_24414", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["governance", "security_gates", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d9ee9d9b1607af9c99a62a6bc1630e4ba79d6c31"}
{"id": "train_24415", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "tests_are_truth", "repo_scale_reasoning", "tooling"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "497cccf68585fe557b0f3a7c9d83d868afa7d416"}
{"id": "train_24416", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "repo_scale_reasoning", "governance", "auditability"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e67a2688833899a76c5502a009a4bcb132a868c4"}
{"id": "train_24417", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "reproducibility", "repo_scale_reasoning", "governance"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5c1bf2a1e6f0be7e7300cc962c9d19c31c07d358"}
{"id": "train_24418", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["governance", "documentation", "repo_scale_reasoning", "evaluation_metrics"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "885f5b5dda86ce9f2435f5d4ea3c376f4976d873"}
{"id": "train_24419", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "tests_are_truth", "governance"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8bfbd1941e47de9a3cf87be24a7f3154c4bf02df"}
{"id": "train_24420", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "repo_scale_reasoning", "governance", "tests_are_truth"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "54aa2fb0887910f830308a75c9e15f0c77d556a5"}
{"id": "train_24421", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "tooling", "auditability", "tests_are_truth"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0c29a0db196dbc38c3e4ef0f4f2b914f9c066865"}
{"id": "train_24422", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "cost_latency_tradeoffs", "tests_are_truth", "governance"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0617b7aae9d3c963a6ff255efb7ccbe7304ac2a0"}
{"id": "train_24423", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "auditability", "tooling", "tests_are_truth"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "14666bbecc61fb1426a36ef6e7c15acd36b446c9"}
{"id": "train_24424", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "security_gates", "auditability"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5ea323a656368ce5e7559e1a0a1fed9991b71108"}
{"id": "train_24425", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "evaluation_metrics", "security_gates", "governance"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a258093e2c3c4cceabf786b4164ee6a4a02396b2"}
{"id": "train_24426", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "tests_are_truth", "evaluation_metrics", "governance"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "81a0a13370f4d067990eeb94ae3261899f1e16f8"}
{"id": "train_24427", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "governance", "repo_scale_reasoning", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3edabc71f088094e4fbcda37e1c475c159c5d9e0"}
{"id": "train_24428", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "security_gates", "ci_integration", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "774355c0f6495842ff857094a3bc273f0b59338b"}
{"id": "train_24429", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "tooling", "auditability", "ci_integration"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f00f27cada86c31fff12b283390a8f7a7b04424a"}
{"id": "train_24430", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "governance", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "53edc0a0ad26b76d214e62aebee09506b34ece6a"}
{"id": "train_24431", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "auditability", "tooling", "security_gates"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a5c1a13dc3cd6cb61d56b37410292057d1da130b"}
{"id": "train_24432", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "evaluation_metrics", "auditability", "documentation"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1c75cf6fd47a14fe223c6d895da8fd7607a1639f"}
{"id": "train_24433", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "ci_integration", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e2902daa1b4dcc91436e83a8fbb3afb5e9de4866"}
{"id": "train_24434", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "governance", "evaluation_metrics", "security_gates"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5572941684c3d53d9832f273f14e63b241777845"}
{"id": "train_24435", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "tooling", "ci_integration", "governance"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1be6c173af10422aac72eeec9a9ea5fb17333863"}
{"id": "train_24436", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "tooling", "security_gates", "governance"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ad6275c44033823773a2dd7f90a4a2ec5689a2cb"}
{"id": "train_24437", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "repo_scale_reasoning", "auditability"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4f0bd8e8558086f931e6fe0fe236672ae03dbace"}
{"id": "train_24438", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "documentation", "evaluation_metrics", "reproducibility"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ea4e116f17623d329d2fd6b42da6a746c6123006"}
{"id": "train_24439", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "evaluation_metrics", "cost_latency_tradeoffs", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f06c138e22dbd4cf20410a71dc024166cfad01a2"}
{"id": "train_24440", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["governance", "tooling", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e6544cad276c07e214ed4d1eceda9b4597e294b8"}
{"id": "train_24441", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "evaluation_metrics", "ci_integration", "security_gates"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3c982d3fc7f1f618ca19c7b0d2e726e58ffa5d59"}
{"id": "train_24442", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["governance", "reproducibility", "tooling", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9f3085396c5f25ecefaceb4ca4d6d6ce7abed9be"}
{"id": "train_24443", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["governance", "ci_integration", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "43c0888eb486dd67e7fdbdb04253dfa2ea12117a"}
{"id": "train_24444", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["governance", "security_gates", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d2e7b9087f559782fa99f720b4613b1d2b708896"}
{"id": "train_24445", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "evaluation_metrics", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "757af2b70b42c21102197b66ce1f2948c5913701"}
{"id": "train_24446", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "ci_integration", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c8ad315e967e2dc25fd887e4150f832106691c2e"}
{"id": "train_24447", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["auditability", "governance", "tooling", "documentation"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "411d259cf0f327999db9828cd86a8af5bc847dfd"}
{"id": "train_24448", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "repo_scale_reasoning", "security_gates", "evaluation_metrics"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4de4b65480836023245f92f5a13d76238d3c5b96"}
{"id": "train_24449", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "cost_latency_tradeoffs", "evaluation_metrics", "reproducibility"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "19e93cceee186b0de406b386a16bbb23386a39e3"}
{"id": "train_24450", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "auditability", "documentation"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3646a8d8f837be0738e428e7a72c4eaac62dedfb"}
{"id": "train_24451", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "tests_are_truth", "tooling"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "28e73140be7aec4d201b0275123e874044dc74a2"}
{"id": "train_24452", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "ci_integration", "repo_scale_reasoning", "documentation"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0eef96e56c254971a405401645dfa6abd1e132de"}
{"id": "train_24453", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["documentation", "governance", "tests_are_truth", "ci_integration"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cd28ee1aabfef22b254bc37a551e108f23700078"}
{"id": "train_24454", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "governance", "repo_scale_reasoning", "security_gates"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c9b2e23698f62bdf4cfdc18364dce65c8b4f2574"}
{"id": "train_24455", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "evaluation_metrics", "ci_integration"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d9987d3d90900bf525ba8a193133a3bbc0f866bb"}
{"id": "train_24456", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "security_gates", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "52847a43b9689b2ea74e3df7e88b43aa74595711"}
{"id": "train_24457", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "auditability", "evaluation_metrics", "governance"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b03a42f233c2815a065acdb74649cb1b26000954"}
{"id": "train_24458", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "governance", "security_gates", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "421f67df90eb0ebfaf66e9429038a959963d1d7d"}
{"id": "train_24459", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "governance", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b4fc366b93440570e2b2df1d601a2d7d330413a9"}
{"id": "train_24460", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["governance", "evaluation_metrics", "auditability", "documentation"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "228f568c0f5c4f7098cc5bc80039f810f895d44e"}
{"id": "train_24461", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "reproducibility", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6268cc4f1d8cae312b2534aeb426b5dd24bc434e"}
{"id": "train_24462", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "reproducibility", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "94fe24a1c4ffc0d22bd872230e2a2f74fc58339c"}
{"id": "train_24463", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "auditability", "documentation", "tests_are_truth"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "71754a4ebc2e0ec455173a621912bce363c40d49"}
{"id": "train_24464", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "auditability", "governance", "security_gates"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f8a5526c1ed8705f1deeacaf17f0275390adc804"}
{"id": "train_24465", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "ci_integration", "auditability"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2fbc4765fa3c2153d74d53cdd5f4565410809356"}
{"id": "train_24466", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "security_gates", "tooling", "ci_integration"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "aadd17fdc36d49c4f794117dae6f3019b39df5fe"}
{"id": "train_24467", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "repo_scale_reasoning", "tests_are_truth", "governance"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6e716811a15bb8ea27ffb050c60db85ddffba48b"}
{"id": "train_24468", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "tests_are_truth", "governance"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "27b20287809be0d9d5f5074a855d02ef91b91215"}
{"id": "train_24469", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "evaluation_metrics", "ci_integration", "governance"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bad2151398b082ff7cf92798df8e89b1e91ba8a0"}
{"id": "train_24470", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "documentation", "repo_scale_reasoning", "auditability"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9e528dafeb58b56aae093551b7cf5e16046ad083"}
{"id": "train_24471", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "auditability", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e3c2d6b113a7bd685a66a3c6916c1f54499f92fc"}
{"id": "train_24472", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "tests_are_truth", "repo_scale_reasoning", "documentation"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3a94ef840695eb3109010046840ce7c24e7b5e00"}
{"id": "train_24473", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["governance", "repo_scale_reasoning", "ci_integration", "documentation"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fb57016ea219d7eded6ba98a3d8c6558f0750f89"}
{"id": "train_24474", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "documentation", "ci_integration", "governance"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2edf0d259227290dac65b1c1145a62173bc36fea"}
{"id": "train_24475", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "reproducibility", "documentation", "auditability"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b986ab7daa8e996fd2bbe0b249bef809f97a5bcd"}
{"id": "train_24476", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "evaluation_metrics", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "500053cea39fa7057c14f56c10cfbcdee84e6d98"}
{"id": "train_24477", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "tests_are_truth", "auditability", "reproducibility"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "013580cf1e6ec798d7eb5020fc54d90d6f39938f"}
{"id": "train_24478", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "governance", "reproducibility", "security_gates"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "aa43501824d4bf9a9e7bf964f1e591b427c8c5df"}
{"id": "train_24479", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "security_gates", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "72357dd750fbb90ce1ecbba802a1cad1665a2dfa"}
{"id": "train_24480", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "reproducibility", "evaluation_metrics", "auditability"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "eb1ddd399277687f9948d266fafdf9e2ae3a74bb"}
{"id": "train_24481", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "documentation", "reproducibility"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d0e406bec463b9107731361268b4cd713634803a"}
{"id": "train_24482", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "documentation", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "73fce57081cdf2ffdc0ea156ab5756fa608b133f"}
{"id": "train_24483", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "tests_are_truth", "governance", "reproducibility"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4c01d767bf25bca92e2e0a00639fcee27df00bf5"}
{"id": "train_24484", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "repo_scale_reasoning", "ci_integration", "governance"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4032351d0072bf16bf14b22270144b522d32b0e5"}
{"id": "train_24485", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "tests_are_truth", "governance", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0f338196aaf69a94ffa70b9ff46265536c642a26"}
{"id": "train_24486", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "tests_are_truth", "governance"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "aea24dec06e9bc8c861d65979e256bc7c28c2824"}
{"id": "train_24487", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "ci_integration", "evaluation_metrics", "auditability"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9e771e3b04a328de646b32b3ca903c3fadfbc556"}
{"id": "train_24488", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "tooling", "security_gates"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3fa02caebb88c87865a440c622a68fd3829e9368"}
{"id": "train_24489", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "repo_scale_reasoning", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cdcb4ee342c5833f1d54c9d1e9c2bb12afcfa88e"}
{"id": "train_24490", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "tooling", "documentation", "tests_are_truth"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b47efe4e499a2cc3f76e4deaf49cf0118350f4f6"}
{"id": "train_24491", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "governance", "repo_scale_reasoning", "security_gates"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "775554de9a4e6d4a3589555a81b6ea3c9f056dc5"}
{"id": "train_24492", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "reproducibility", "ci_integration", "documentation"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "633bf47d005675bbb0350b69a34b4e1d5d834b2d"}
{"id": "train_24493", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "tooling", "security_gates", "auditability"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7203cc3cef03f179953ce5762232fac758da566c"}
{"id": "train_24494", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "tooling", "ci_integration", "security_gates"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b8bce8e2faceefb423e542bc342e5813387d7fe7"}
{"id": "train_24495", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "evaluation_metrics", "ci_integration", "security_gates"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "de050614581577988242c078a91ab61c0208deab"}
{"id": "train_24496", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["governance", "cost_latency_tradeoffs", "repo_scale_reasoning", "tooling"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5bdd028724f41f397e33a527545d6bd04e89c619"}
{"id": "train_24497", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "governance", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "da15d1bfd9c885e2b09e2e99b8b7e21b01e0c011"}
{"id": "train_24498", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "governance", "ci_integration", "repo_scale_reasoning"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2e42b09f292af5cb8bf8fddb0e8d1045ab45c2cd"}
{"id": "train_24499", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "repo_scale_reasoning", "tests_are_truth", "tooling"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d09e2ebba92e123d6ce6068dbcb7deeef30d5da4"}
{"id": "train_24500", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["governance", "cost_latency_tradeoffs", "auditability", "repo_scale_reasoning"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d170cbb1ad14ee4b6879d36934af3b4d32735435"}
{"id": "train_24501", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "tests_are_truth", "evaluation_metrics", "reproducibility"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4b4e4985589d1a4236d581645a18c6a5716488ba"}
{"id": "train_24502", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "repo_scale_reasoning", "tooling", "documentation"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "25147ed019fb8d11477f367bfcc69dec24fc0418"}
{"id": "train_24503", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "reproducibility", "governance", "evaluation_metrics"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "709ace0cfd04b135c4c3469f97ea8bf815a6c1f4"}
{"id": "train_24504", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["governance", "tooling", "auditability", "ci_integration"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "addaba041ef733384232e45bfbbf33ce46cc6ab7"}
{"id": "train_24505", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["documentation", "reproducibility", "tooling", "evaluation_metrics"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9ce17238cb4e00d87940a904e4d5646f0b416a25"}
{"id": "train_24506", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "ci_integration", "governance"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d0a90f08bd6dd83e4ddd15b06ddf19ccc65e673d"}
{"id": "train_24507", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "tests_are_truth", "cost_latency_tradeoffs", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "adbcef64f2380b65e518fca5eedbf876d3c9e44a"}
{"id": "train_24508", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "ci_integration", "tooling", "governance"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7bbdd9dd2d43f45d646cdd431b2d24af8eddbf8f"}
{"id": "train_24509", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "reproducibility", "documentation"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7993b00587c4c47639093db1e8fe2211b8f88c06"}
{"id": "train_24510", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "documentation", "auditability", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "05b11a494abd9b23c407d3a2a36c0a4074b9f513"}
{"id": "train_24511", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "reproducibility", "auditability", "tests_are_truth"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a5811a87d1693bcb8e9441aeb5b482d390c27566"}
{"id": "train_24512", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["governance", "security_gates", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1bcddf25419b5070c134cb329fda83c006f9ef84"}
{"id": "train_24513", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "tests_are_truth", "security_gates", "tooling"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "675b24de745a0b477bd163a17baf60e9f2d966a2"}
{"id": "train_24514", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "repo_scale_reasoning", "security_gates", "reproducibility"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d0a8c71d9c828ee78142dafb26097f29171c24df"}
{"id": "train_24515", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "governance", "cost_latency_tradeoffs", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "01c6a988f94d03b97450b4059c38940ecd5872a3"}
{"id": "train_24516", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "reproducibility", "evaluation_metrics", "tooling"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cff8ead94eec108b63238a06716d6c73722cbae0"}
{"id": "train_24517", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "governance", "auditability", "documentation"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0b694d87f03ec14b8d0691f3b23e92cc4644f0c6"}
{"id": "train_24518", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["auditability", "governance", "ci_integration", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3f74bce55dc67cee69ea1246d53da2601dc04c16"}
{"id": "train_24519", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "auditability", "cost_latency_tradeoffs", "governance"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "82165521e798a6c74ba1c6bba49acd35b5b95b2f"}
{"id": "train_24520", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "governance", "repo_scale_reasoning", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "efc4d3089e5e1681345e793d573329314fd8ad0f"}
{"id": "train_24521", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "tests_are_truth", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8b38a5ae8ee82515bb9de43661dd57f7727c5388"}
{"id": "train_24522", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "security_gates", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d3367a99423e78c6da7399c3d3e1820dfada8246"}
{"id": "train_24523", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "auditability", "tooling", "documentation"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "62750cfaefab5f99033dd1bb38cee07855f4cf56"}
{"id": "train_24524", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "repo_scale_reasoning", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7a83052652553942dc8cd854249fa8c3bd6a0ca3"}
{"id": "train_24525", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "governance", "tests_are_truth", "security_gates"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8acb011c8aaf7021c42a76bc2616a81897e1d83e"}
{"id": "train_24526", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "documentation", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "63667a6a749b967300c96f5e2b905536fa8a977b"}
{"id": "train_24527", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "security_gates", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "62df666d82a9df7dd3d5541e44446865a33cb326"}
{"id": "train_24528", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "governance", "evaluation_metrics", "auditability"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7b07d810b8fa60cd15f0e869d95c387a4d504e9d"}
{"id": "train_24529", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "tooling", "governance", "documentation"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "50272b8ed0efbaa9d47eb27a1a0404ccf4dd1f90"}
{"id": "train_24530", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "reproducibility", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ed07d2af8b41133b143c0c53601b417bf3d54d96"}
{"id": "train_24531", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "tooling", "ci_integration", "security_gates"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e171b6dd610e0b47499b83c1c85fb70a5126e26b"}
{"id": "train_24532", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "ci_integration", "auditability", "tests_are_truth"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e97a30ce647a6850298a885e6c28234b5513c54f"}
{"id": "train_24533", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "reproducibility", "governance", "documentation"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7b33f34c9c8b5f88b1e477053800592ef4bfa7b7"}
{"id": "train_24534", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "auditability", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f98444a4f4086336fa30496bbc4714c539455fff"}
{"id": "train_24535", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "governance", "ci_integration", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c134c800b34df18dbb02b447b2b73dd217ce8897"}
{"id": "train_24536", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "auditability", "tests_are_truth"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d5291c4b98de3240c350177685feb2f044103797"}
{"id": "train_24537", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "repo_scale_reasoning", "ci_integration", "evaluation_metrics"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f4fe0986c87372f51cea0ad78d2014ac6ff5aff6"}
{"id": "train_24538", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "security_gates", "tooling", "auditability"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d57ef54b77b9abc2bda82bc02be12dd3c4e106a8"}
{"id": "train_24539", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "evaluation_metrics", "reproducibility", "security_gates"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0919a29c9ba0fedf5b465ddbc6eb0f3917cd8fa6"}
{"id": "train_24540", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["auditability", "tooling", "ci_integration", "security_gates"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "040ac6e74424bbf17b0d7a63ebd6ad2d4bcaddf7"}
{"id": "train_24541", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "reproducibility", "ci_integration"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2e60b7c25c33bb805c2d5cdf0a135f43e3fea1ff"}
{"id": "train_24542", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "governance", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4cb2c3c10e020f3c42b270a6f4f4f67f2736b1b8"}
{"id": "train_24543", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "repo_scale_reasoning", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6827b80e788eaeaecb1202a535c7ab3285ccd7ad"}
{"id": "train_24544", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "reproducibility", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "64697bf40bd97dab151d6ece41c16e1932128c8d"}
{"id": "train_24545", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "repo_scale_reasoning", "ci_integration", "governance"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "96d4fdff6527549f2d823c7998186a1a3fbe11d0"}
{"id": "train_24546", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "tests_are_truth", "repo_scale_reasoning", "tooling"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0cd68bcae2c5472cd343ee3e54ba9e99a4528d9f"}
{"id": "train_24547", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "auditability", "tests_are_truth", "governance"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d6fa74bd7fd277bfcaf0ed3235be1842e43396d5"}
{"id": "train_24548", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "repo_scale_reasoning", "tooling", "auditability"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c4617ab8451239592564ecdd6d79a0f936a71df6"}
{"id": "train_24549", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "documentation", "security_gates", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e7f78baadf41d8d3d0ec6c16b0727c10cf9b187a"}
{"id": "train_24550", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "evaluation_metrics", "governance", "tooling"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3d74931321744b25fecff8b48cb56f4857b3620a"}
{"id": "train_24551", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "documentation", "auditability"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "830d7780e23d7210c3d69c8379a005e52af9a516"}
{"id": "train_24552", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "evaluation_metrics", "security_gates", "tooling"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f83e466f624fd04ba57f76df2875004ec58c3b54"}
{"id": "train_24553", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "repo_scale_reasoning", "tests_are_truth", "security_gates"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "840d7eb46c90aa7c7e973b98cf6e824771861fed"}
{"id": "train_24554", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "security_gates", "repo_scale_reasoning", "tooling"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "13512fa131180ac60930360cdb4a1e7cf4483b42"}
{"id": "train_24555", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "evaluation_metrics", "ci_integration", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f790ce24fdd29fa1b32f35027c700d0bcd89064e"}
{"id": "train_24556", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "evaluation_metrics", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "636c0ce3c826ec6a155ed8b66426c50fcf9fb6b5"}
{"id": "train_24557", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tooling", "governance", "auditability", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "112c9fc736b622eb1851c306f57b4a5ae0578c29"}
{"id": "train_24558", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "evaluation_metrics", "governance", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3627e49bd3f0695b169ea29024ded5f1bc31eafa"}
{"id": "train_24559", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "governance", "tooling", "documentation"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f4e9a8211cf4ee3fff9e0e8b884ad8c6fc6185f9"}
{"id": "train_24560", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "repo_scale_reasoning", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "acab0c0f52789f93190bce4a3096b4b99f78b88d"}
{"id": "train_24561", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "evaluation_metrics", "tooling", "security_gates"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7ad015aa7774e8855a8012671d5cd28dcfffff15"}
{"id": "train_24562", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "documentation", "tooling", "security_gates"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3892654b1e2fd2ac02efac3e8da99ae108438e38"}
{"id": "train_24563", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "governance", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "beb0c88644d0829b65e069c641f30f7866b9a76d"}
{"id": "train_24564", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "governance", "tooling", "reproducibility"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d111a7dc50cb88f38f3ea60783671fc1ef765243"}
{"id": "train_24565", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "ci_integration", "reproducibility", "tests_are_truth"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cd888dafde661b115ce6e626856180369a031866"}
{"id": "train_24566", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "auditability", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f5ca31cd3bc43de20aaff70be64426c970d7a8d1"}
{"id": "train_24567", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "governance", "auditability", "evaluation_metrics"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "931322d5c22cdbdb5db412d784b6317a942fec89"}
{"id": "train_24568", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "documentation", "ci_integration", "reproducibility"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9f40dfaa4f47b3c690dd7314bd60fe76b661845d"}
{"id": "train_24569", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "governance", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "88e3ac67cfd1ff77097905a53a8160ae379d9bd2"}
{"id": "train_24570", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "ci_integration", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "752a044ac99ed40945fb0b7ffc9c2ca3160f5bc0"}
{"id": "train_24571", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "governance", "ci_integration", "security_gates"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fd61884349c47211f41fa703220fc1cd94a1f02f"}
{"id": "train_24572", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "ci_integration", "governance", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "757cf394289043792c3830f0aee58d717299bdcc"}
{"id": "train_24573", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "documentation", "tooling", "evaluation_metrics"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "34c5122b6ea3ebf4d89d152ec30bb3b0ba1b3602"}
{"id": "train_24574", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "security_gates", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f7d4af82b09146ba99a269ebca8bb786c86efecd"}
{"id": "train_24575", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "reproducibility", "documentation", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ca458ba4318cc949c44864500e70bdff3bf77cd0"}
{"id": "train_24576", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["governance", "documentation", "tooling", "tests_are_truth"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6d58267d912aafa470526078f9fc47a970c923d4"}
{"id": "train_24577", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["documentation", "tooling", "security_gates", "reproducibility"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "34416546579f1149b724e5879670c2363ee988d7"}
{"id": "train_24578", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "auditability", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5a0a0215a163b0be574858bcdfd0dcbb0ac13459"}
{"id": "train_24579", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "documentation", "reproducibility", "security_gates"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2cd07960103df59b4fbf0992e825c1ee986d5fa4"}
{"id": "train_24580", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["auditability", "ci_integration", "governance", "reproducibility"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fb0accc78c6ff598d18f8ac0fad84e2b336753d2"}
{"id": "train_24581", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "reproducibility", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "042ebd0002c0501397376d40132beda486cbe319"}
{"id": "train_24582", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "documentation", "ci_integration"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "28335e8910c5f5a7b8b1e1785fd736f639ce0e22"}
{"id": "train_24583", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "auditability", "security_gates", "ci_integration"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1d3bed88d17fc2d30ca2398ad7c23ce4d0ac6a80"}
{"id": "train_24584", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "cost_latency_tradeoffs", "security_gates", "tooling"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a41236f227c62de0f2755257f704a22469a32790"}
{"id": "train_24585", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "tooling", "auditability", "governance"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "192155851437f1d1d3dc80f08dc012e92c484b2e"}
{"id": "train_24586", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["governance", "ci_integration", "reproducibility", "evaluation_metrics"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cce404efd0dd2e91f1a2158b4485a88055c352a6"}
{"id": "train_24587", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "evaluation_metrics", "cost_latency_tradeoffs", "security_gates"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "88c77271b0341c6d190f8f20015c7241c430654b"}
{"id": "train_24588", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a5899f28beef77105a94d8829128f7a6b6f06bc7"}
{"id": "train_24589", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "reproducibility", "security_gates", "governance"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6ba5cbeac014a018f280276b48d74c4911e0f51b"}
{"id": "train_24590", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "auditability", "repo_scale_reasoning", "tooling"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6521b558d41500123ff3b96c1865dbe3d2d6cad1"}
{"id": "train_24591", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "evaluation_metrics", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "37b77d0750f355970b2859b483ee1bcb5a52a77f"}
{"id": "train_24592", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "security_gates", "tooling", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cbc14cd93f18b2e2034f15c7a301baf4f2864808"}
{"id": "train_24593", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "tooling", "documentation", "security_gates"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2d7dd1021a3535f88be465a5960081ecfc8f5da5"}
{"id": "train_24594", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["governance", "documentation", "reproducibility", "tooling"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "89429d6948ed68f726df74b29afce98471d1dd5b"}
{"id": "train_24595", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "tests_are_truth", "tooling", "reproducibility"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d06bbdb533a1b899631b1ab1d39b2dcb6f2cb043"}
{"id": "train_24596", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "evaluation_metrics", "governance", "documentation"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "483baefe1d841e70dd95be5b730ddbe8abdc1737"}
{"id": "train_24597", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "security_gates", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3613720c041a2a20ea97da6887c030ee7f5716ce"}
{"id": "train_24598", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "tooling", "evaluation_metrics", "tests_are_truth"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8bda9a0919de3a90887adccd9eef80fd02997eac"}
{"id": "train_24599", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "documentation", "ci_integration", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d63bac7f04867d5de5bd1cd5b35b712fbadb79e3"}
{"id": "train_24600", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "auditability", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9c0c87a0c84b0303053c7daa8027f120518ae507"}
{"id": "train_24601", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "security_gates", "reproducibility", "documentation"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1982e3212169d7f73799960cfe7474690752b5da"}
{"id": "train_24602", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "tests_are_truth", "ci_integration", "security_gates"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c514161a197ce05a31853cc883704cfd7d394509"}
{"id": "train_24603", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "tooling", "cost_latency_tradeoffs", "repo_scale_reasoning"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cfb5e82064d9c11b1be22d575debf2cb7cf8a61c"}
{"id": "train_24604", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "auditability", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6775dafb5136ebf7b3a1babbbe894954c9ea5fd9"}
{"id": "train_24605", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["auditability", "documentation", "evaluation_metrics", "security_gates"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4aef6afcd8395a42d22e72f46af1aac04e88bc0f"}
{"id": "train_24606", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "governance", "evaluation_metrics", "documentation"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dd22fb91beedd626991fa006eb86b5f491efebe9"}
{"id": "train_24607", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "tests_are_truth", "tooling"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cf286160ff90b6e4df305d3c661f6ef64a1ea627"}
{"id": "train_24608", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "documentation", "auditability", "tooling"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "011ca5431e7a6c5f8b7801361640b6000db0a035"}
{"id": "train_24609", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "ci_integration", "documentation", "tooling"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "314ed10ff5485d1b4a7945291aa669e4608bc3f4"}
{"id": "train_24610", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["governance", "documentation", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0a2dd49e4982d65c2e0f426227b18658233e8f1f"}
{"id": "train_24611", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["governance", "tooling", "auditability", "security_gates"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6f0de66bfd4dca2d83d6552089b202932dfb65bf"}
{"id": "train_24612", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "governance", "repo_scale_reasoning", "evaluation_metrics"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c4fa27e35fa8471e93d6bb9a2d1315097f04a090"}
{"id": "train_24613", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "auditability", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f86599ff0c480d2530546b36911fa44246080dd1"}
{"id": "train_24614", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "governance", "auditability", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "86196e774e155ad2e08d6831fa0dec4e1b1e8748"}
{"id": "train_24615", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["auditability", "reproducibility", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cb25fd3209568c33f479dce1676e9c49dd693bdc"}
{"id": "train_24616", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "reproducibility", "governance", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "25f88e1a689371d20e1af11e44e7e697a7594838"}
{"id": "train_24617", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "auditability", "ci_integration", "security_gates"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a19095ce24a4610417aba3096103c4416fb0dfa3"}
{"id": "train_24618", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "auditability", "documentation", "security_gates"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "23a1a42b97b53afad135f5e65529298b0bc9c9fd"}
{"id": "train_24619", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "ci_integration", "reproducibility", "auditability"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ba914c3b2dcb00a5b232282865bad9166307df4c"}
{"id": "train_24620", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["governance", "auditability", "ci_integration", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dbb870640d2e1b10c5ae15b91fab80d7fc9f4f07"}
{"id": "train_24621", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["governance", "auditability", "ci_integration", "tooling"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "112ff6ad8476e8e954ab8ffa67f24bf7c4a4b14b"}
{"id": "train_24622", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "governance", "documentation", "security_gates"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d6debb5cbd3657b50c3187283cb91310bb6cb04d"}
{"id": "train_24623", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "reproducibility", "governance"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a19748fd5b544cd1025c47de091c2c66dd421190"}
{"id": "train_24624", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "ci_integration", "evaluation_metrics", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c733066620bccf774786735df7e83f5447561421"}
{"id": "train_24625", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "evaluation_metrics", "tooling", "tests_are_truth"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7060635b2621a39379fa9c889bcb35cc846e3d14"}
{"id": "train_24626", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "auditability", "governance", "reproducibility"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ee158e9bb24006bc9464ebd6d7357c71612aa63c"}
{"id": "train_24627", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "repo_scale_reasoning", "governance", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "405838758f38f1b8b4e6a81932a5514b316baa77"}
{"id": "train_24628", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "security_gates", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "773d996c0a63b2302f4d475cb296d81d8b4d4a79"}
{"id": "train_24629", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "documentation", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "740738bc65380e40f1feb46ef7b3c11576e17f3f"}
{"id": "train_24630", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["auditability", "cost_latency_tradeoffs", "evaluation_metrics", "reproducibility"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "49a8b50428f0d6145c82e122e39d73d75adafd01"}
{"id": "train_24631", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "auditability", "tests_are_truth", "ci_integration"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c678188630228e2a0096570432ff8a0519012ec4"}
{"id": "train_24632", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "ci_integration", "security_gates", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1b63ffafbb4aa0426dacb4f449f8cfb7a9f22b2e"}
{"id": "train_24633", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "reproducibility", "auditability", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7e6898bceec258d36069737257012d3f3f1ba8de"}
{"id": "train_24634", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "tooling", "repo_scale_reasoning", "documentation"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "21aa34be1a634c8254bec4795a7a6de1ab71e369"}
{"id": "train_24635", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "auditability", "evaluation_metrics", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2669d0e5763b5ed41a8640b5eda1a49d5e19a854"}
{"id": "train_24636", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "governance", "documentation", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0c52ceb99b61ddfb1676fcc6cfe43029bde4f4c5"}
{"id": "train_24637", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "governance", "tests_are_truth", "security_gates"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c0b73e1c4722443f07da0f5ecbd5d1dd9371ca7e"}
{"id": "train_24638", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "ci_integration", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f055e7be1132461e848ca088be52e8ab64c9e99e"}
{"id": "train_24639", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "ci_integration", "reproducibility"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7bf9c5e01d2b1c03bac51612de19a7767d921537"}
{"id": "train_24640", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "reproducibility", "evaluation_metrics", "tooling"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "16b1597c60148d1d6a86865bfc771ccf95f58500"}
{"id": "train_24641", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "reproducibility", "cost_latency_tradeoffs", "governance"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "75c4b768003bc8dd56d79ebeb5f13f8e5b898262"}
{"id": "train_24642", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "security_gates", "tests_are_truth", "documentation"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9ca8ee6a4787745cf1693955ac040ca507148770"}
{"id": "train_24643", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "governance", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "569f673eb1ed65a454628b632fc0c7f0e7f04d74"}
{"id": "train_24644", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "tests_are_truth", "tooling"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4a71012c773da3bf621ee832ca8c7d75528802c3"}
{"id": "train_24645", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["documentation", "evaluation_metrics", "governance", "repo_scale_reasoning"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ee042074a6cfb87dc19cfc6501941ada64539cd6"}
{"id": "train_24646", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "tests_are_truth", "auditability", "reproducibility"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "651527682d2c7a94bdeb1777098c7c73ab939849"}
{"id": "train_24647", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "tests_are_truth", "documentation", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "92f1e200c1ba9e3c1523b19645e2229283ad691a"}
{"id": "train_24648", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "tests_are_truth", "documentation", "governance"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8534ae78ecf4e49327adec448f8952a924586265"}
{"id": "train_24649", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "reproducibility", "governance", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4b67cc9b6a425ece1a374db13ba9034341ac6c94"}
{"id": "train_24650", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "tooling", "security_gates", "tests_are_truth"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7d369cd4c939005371368612fbeab6ce112bbd02"}
{"id": "train_24651", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "repo_scale_reasoning", "reproducibility"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3106c5fc4385724208bf7555a10318c42aeb1d8c"}
{"id": "train_24652", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "ci_integration", "tooling", "security_gates"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0033e5e147835f5589212bde897faa08d6b74875"}
{"id": "train_24653", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "tests_are_truth", "documentation"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "85300eb8d0b92c0fd2d09864d73bda1ddae4daeb"}
{"id": "train_24654", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "evaluation_metrics", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dd05373b10fdbc317be5b5b2420c8a362b36cfe7"}
{"id": "train_24655", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "evaluation_metrics", "reproducibility", "ci_integration"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7bf9229964b992a21646f7b6f488d6e9202c6427"}
{"id": "train_24656", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "governance", "auditability", "reproducibility"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7165ffdf2d0915b2ede3cdc01e12a7e90d9d39d1"}
{"id": "train_24657", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "ci_integration", "governance", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f32f7a49fafde691a0b2aa45f782453b4ff85f60"}
{"id": "train_24658", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "repo_scale_reasoning", "tests_are_truth", "documentation"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f4566ae28e82a134f1bfb38e382433079b1459b5"}
{"id": "train_24659", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "tooling", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "07cad43701b999da8538a7506ada6893088df76c"}
{"id": "train_24660", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "tooling", "auditability"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "58f4419d58b3045e5593a2578a96a6e67f1e1078"}
{"id": "train_24661", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "tests_are_truth", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4a1ce4ef0088125a37ffaec943bdb92d6a0b2eb8"}
{"id": "train_24662", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "tests_are_truth", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "600f35a9d938a785358459ae24839e935936ecc8"}
{"id": "train_24663", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "security_gates", "tooling", "auditability"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0c7a2e0219992dea3a422a90aa2e111805c0cab3"}
{"id": "train_24664", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "tests_are_truth", "auditability", "security_gates"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c092d4560166ec70ee57badf5e6b4b54f8a220fe"}
{"id": "train_24665", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "reproducibility", "tests_are_truth"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "440aa4a0639c7c385f9387bfefe0133f3e68bff6"}
{"id": "train_24666", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "reproducibility", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "27d1b37b67ab1fc8745da5ac5f9c823ec28e002f"}
{"id": "train_24667", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "security_gates", "repo_scale_reasoning"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cb4a7506a52497c28c1ec9aad27b5e2115ca49cf"}
{"id": "train_24668", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "auditability", "ci_integration"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9a6042ee1b5c965c154c33b93f099762748cc224"}
{"id": "train_24669", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "auditability", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a8c5b9d0e49023e5e8c19307847b7e2a13814d5b"}
{"id": "train_24670", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "documentation", "governance", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a7f37bca45abb139e8daf3e674fdaf7bd8747abc"}
{"id": "train_24671", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["auditability", "repo_scale_reasoning", "security_gates", "tooling"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3a94ef840695eb3109010046840ce7c24e7b5e00"}
{"id": "train_24672", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "documentation", "auditability", "ci_integration"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "535e1430a648b7e7e42bbc7c1eb057f66e0b8d54"}
{"id": "train_24673", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "governance", "documentation", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "684f08c84d391c0b834f50b7e582c9239204471d"}
{"id": "train_24674", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "security_gates", "repo_scale_reasoning", "evaluation_metrics"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9fffcfe5c1008244efa216da47949ac6edae7e8d"}
{"id": "train_24675", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "governance", "documentation", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8a5423890e3be871a44bca7138b3b4b5302f82b4"}
{"id": "train_24676", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "tooling", "auditability"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2be4c8065e451ae95b8f8321753f75a550a34b4b"}
{"id": "train_24677", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "reproducibility", "auditability", "repo_scale_reasoning"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ff99bad3f45a21f23ff3fe2b7703bf9bd37065c5"}
{"id": "train_24678", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "reproducibility", "security_gates"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "17c494260c3000beca4dff6b07df4c9de394e755"}
{"id": "train_24679", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "documentation", "ci_integration", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "77dd33830800edf6138b0fee69b5592c3d03701a"}
{"id": "train_24680", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "documentation", "repo_scale_reasoning", "auditability"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "610cbe87dc93811be6601f713fb3ef0dc36b3c11"}
{"id": "train_24681", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "tooling", "security_gates", "documentation"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "624aec2566fe43938c5953a7768401ea74793858"}
{"id": "train_24682", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "security_gates", "auditability", "governance"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "51d78f19d1f0a01277dc108a75ec6abed7625848"}
{"id": "train_24683", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "tooling", "repo_scale_reasoning", "evaluation_metrics"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7bf282ab27819eb72b91127540cf2a6114c29765"}
{"id": "train_24684", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "security_gates", "repo_scale_reasoning", "governance"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1d71a047391c5fa4212185260641d03e284c1239"}
{"id": "train_24685", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "ci_integration", "documentation", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "467906217fc7580b801f5a0f1d173779daf9e4df"}
{"id": "train_24686", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "repo_scale_reasoning", "documentation", "security_gates"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "838847b678bb9a097cb4ad91188e85d54e4f730b"}
{"id": "train_24687", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "tests_are_truth", "tooling"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e31cd7b686876c72a3ed32d7a4f81fe2739ee5f4"}
{"id": "train_24688", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "reproducibility", "auditability"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d44110c63f93c3387d6af4b66d9237e75771f61e"}
{"id": "train_24689", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "documentation", "evaluation_metrics", "reproducibility"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1b2206f0326450f8b31f33d15c791132056c67e7"}
{"id": "train_24690", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "security_gates", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "51e0d8e56bce643faeaa51670ad6377293168503"}
{"id": "train_24691", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "documentation", "auditability", "security_gates"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bb0a1d9c2ed1552b804af14e81838cfd6e64e8af"}
{"id": "train_24692", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "evaluation_metrics", "documentation", "auditability"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2473f51694784fd2f1185fe40aea3764ebf5d0cb"}
{"id": "train_24693", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "governance", "security_gates", "documentation"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "be9d39e5c22ffd01ef3caae4ffbb75b6800af25f"}
{"id": "train_24694", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "documentation", "cost_latency_tradeoffs", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cb2b42cc9168a637399d8cb8dca888128366d449"}
{"id": "train_24695", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "evaluation_metrics", "tooling"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "79df8183b465d9f5cf37ec3c076e8ca68e926689"}
{"id": "train_24696", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "security_gates", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "579fcb57eefbfd081a5dbfcfb6aaa16648654f47"}
{"id": "train_24697", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "governance", "repo_scale_reasoning", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "23fd7042780392ebbc1584ab50a5ea1d5bb1c545"}
{"id": "train_24698", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "security_gates", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "191f89d51ddafa2fc97d018445ba00133b48aca0"}
{"id": "train_24699", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "governance", "repo_scale_reasoning", "security_gates"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "41c1c9781c9c55f0c5e57f8b105469ce45c9a347"}
{"id": "train_24700", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "governance", "auditability", "ci_integration"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "df424611e55c084e4e26252996c61b398a7c426b"}
{"id": "train_24701", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "repo_scale_reasoning", "tooling", "reproducibility"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "98213f13539b4d478cd965912d8b5f46715f948d"}
{"id": "train_24702", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "governance", "security_gates", "auditability"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3b631f1e79c51c1b23cf15b678c2224bc25da9f3"}
{"id": "train_24703", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "tooling", "ci_integration"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f03cb5a0bca0e8872310903a6c948d64ca40b945"}
{"id": "train_24704", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "auditability", "tests_are_truth", "tooling"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "db9bf85c3182bf2e31c13c62def526feea5517a5"}
{"id": "train_24705", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "governance", "evaluation_metrics", "auditability"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c4ef79b5b420626c37f8a020915993a9d3af648c"}
{"id": "train_24706", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "reproducibility", "auditability"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "18b769580000862c41551c91bde652c05fec70cd"}
{"id": "train_24707", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "ci_integration", "security_gates", "evaluation_metrics"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "166d2b9329d42404129d039839b7d14dc12febf3"}
{"id": "train_24708", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "ci_integration", "security_gates", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "42fdff52ed6c870ece44d16c5adb3ff801e95d10"}
{"id": "train_24709", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "tooling", "cost_latency_tradeoffs", "security_gates"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c08b39d1de3e3d0d0729d466acd2f12d6e1b39e1"}
{"id": "train_24710", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "tooling", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f417b35c25328b9f4857d2e5cae0f48498903570"}
{"id": "train_24711", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "ci_integration", "evaluation_metrics", "tooling"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d269cfc20b84cb668c561c0910895125398a3d83"}
{"id": "train_24712", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "documentation", "repo_scale_reasoning", "tooling"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "547dcf5c956afcea8d3dc2f097aab6b53357308f"}
{"id": "train_24713", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "ci_integration", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9cdad9748e9d551aa8292379b4260049a58716f4"}
{"id": "train_24714", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "evaluation_metrics", "reproducibility", "auditability"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8380b059d06410050353775b444010da1665ddd9"}
{"id": "train_24715", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "tooling", "evaluation_metrics", "auditability"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cd9fd69273a26ef37e26d2b222c7c440401d64a6"}
{"id": "train_24716", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "ci_integration", "tooling", "repo_scale_reasoning"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d815a0d2889aba0b84d17bf6c67eb3861a602b71"}
{"id": "train_24717", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "auditability", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f1980579b766194c76210a209d90ff3d3b5286ae"}
{"id": "train_24718", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["governance", "auditability", "documentation", "tests_are_truth"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d0cfe039fd3f8b278acdfe123df8e9fffb9c82ae"}
{"id": "train_24719", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["documentation", "tests_are_truth", "auditability", "ci_integration"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c66a291fa24bb3cd0239ccfd321f38cbe3b7cd0d"}
{"id": "train_24720", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "reproducibility", "auditability"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4313b961e34bcc1b0b15fe3aae1c33645b1f58d9"}
{"id": "train_24721", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "ci_integration", "governance", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "46775619661df0e051f30ff4f4f5717cd5ea5f5e"}
{"id": "train_24722", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "security_gates", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a82c53b10dc1cba33804c7ee146cf8e4b03c335d"}
{"id": "train_24723", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e656538733914f1326033da33fbbc92ae9247caf"}
{"id": "train_24724", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "ci_integration", "tests_are_truth"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a7e252071df823f941fdd65de0a6205fc89757ce"}
{"id": "train_24725", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "ci_integration", "security_gates", "governance"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8b56f0c4f71c4c86a35a408e3080161e22dda34c"}
{"id": "train_24726", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "tests_are_truth", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f566b71874b6ba09033e7061cb5de2a1ce2d00ed"}
{"id": "train_24727", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "auditability", "ci_integration"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4dfbc3d902a656694392cccf7aa4c132c8834a55"}
{"id": "train_24728", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "tests_are_truth", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9d008e18e21e62705d8398da94f47dad02bd5705"}
{"id": "train_24729", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "ci_integration", "tooling", "evaluation_metrics"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "028587454e1e70fc03f16ebd05c31f831bede196"}
{"id": "train_24730", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["governance", "documentation", "ci_integration", "tooling"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "723612a182ebb66056eb03f2b949ced5c5ff4608"}
{"id": "train_24731", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "governance", "ci_integration", "evaluation_metrics"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9aa1c4238ecbe12579bda54aaacf81fc3562455c"}
{"id": "train_24732", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "auditability", "tooling", "governance"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b849e3810d540b62087e821ab78f4c8ecf217509"}
{"id": "train_24733", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "governance", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dfcaf467c3724658671fe9ad030f05232940bdc1"}
{"id": "train_24734", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "repo_scale_reasoning", "tooling", "auditability"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "98585192604c9aad53e27b96181cef02ad3d1932"}
{"id": "train_24735", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "cost_latency_tradeoffs", "repo_scale_reasoning", "documentation"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3962af02e39adea8ec43b5247f59382d67cbf95d"}
{"id": "train_24736", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "documentation", "ci_integration", "security_gates"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5c0f757981bf71d47efb5d946499d85313681184"}
{"id": "train_24737", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "tooling", "governance", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7b80834007cdd6e69118417ea9f03159762c041e"}
{"id": "train_24738", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "auditability", "ci_integration", "reproducibility"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "48d33959b0e7d343329e3834f63fe1fbbd9211e4"}
{"id": "train_24739", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "repo_scale_reasoning", "ci_integration", "governance"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bbce43d0ba91ed01e1e0d11312daa691ddd5320a"}
{"id": "train_24740", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "security_gates", "tooling"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "672c74641191544321e4cc08f0f39f4150a9ad03"}
{"id": "train_24741", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "security_gates", "reproducibility", "tests_are_truth"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b8521b47635edce060b880264ddc471949e26afe"}
{"id": "train_24742", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "security_gates", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d4e8334e310790b82a6d3b5bd8da69b55c49e966"}
{"id": "train_24743", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["governance", "auditability", "security_gates", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1d07a0d4da34f275a0c75cf6eccac60d50ed5597"}
{"id": "train_24744", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "security_gates", "documentation", "tooling"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a5f1343d18d3c2590f5d693dacd6369cdb472ebc"}
{"id": "train_24745", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "evaluation_metrics", "ci_integration", "documentation"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "37b77d0750f355970b2859b483ee1bcb5a52a77f"}
{"id": "train_24746", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "tooling", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "07de88a0283d234fafc08c97d840996f2bb3e5ca"}
{"id": "train_24747", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "governance", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "970431286f0083848a424eb3dbb2d80c1cf870d6"}
{"id": "train_24748", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "evaluation_metrics", "tests_are_truth"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fe948ecd77d0c3dfc45baf70296a83b853e78278"}
{"id": "train_24749", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "ci_integration", "tooling"], "moe_experts": ["evaluation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bff3fc7a00f8186a7f360affee1c57fcb61eb597"}
{"id": "train_24750", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "ci_integration", "tooling", "auditability"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9751bca74c81f5d865288b0ca3ef9340b9b0e950"}
{"id": "train_24751", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "tests_are_truth", "tooling", "ci_integration"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "35b34ba03008105a4644eeddaad627d795d26eb0"}
{"id": "train_24752", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "ci_integration", "repo_scale_reasoning", "documentation"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "276d9513d759a559bfd5d6745b8a11edb84bade6"}
{"id": "train_24753", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "repo_scale_reasoning", "auditability", "reproducibility"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c6a4420c3abaf22b63d2e5661d64dbf73c120728"}
{"id": "train_24754", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["auditability", "repo_scale_reasoning", "tooling", "governance"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "124a5a48202361bf05c8e8e16c1abd1b6ed02bd6"}
{"id": "train_24755", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "governance", "evaluation_metrics", "documentation"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1d34db206b4a9c4c2b26e1609c635cba33e178a0"}
{"id": "train_24756", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "ci_integration", "cost_latency_tradeoffs", "security_gates"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2b55e316dda5a14692a5091c1e9fc5f4edd5b589"}
{"id": "train_24757", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "security_gates", "ci_integration", "governance"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0fe6b44f28de1008e8b3cda7a7b3595d225f81bd"}
{"id": "train_24758", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "documentation", "security_gates"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dc002777883df6f63bd1d1efe50aa0348c651427"}
{"id": "train_24759", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "repo_scale_reasoning", "evaluation_metrics", "auditability"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "986fa607ac2986c771a312615e1ee3a734745917"}
{"id": "train_24760", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "governance", "tests_are_truth", "reproducibility"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2ca3457491e93efe720f57c1cab268b37523761a"}
{"id": "train_24761", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "cost_latency_tradeoffs", "evaluation_metrics", "tooling"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7968ebfe77f2daa5250fe5c7816c4d1dbbeeb552"}
{"id": "train_24762", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "reproducibility", "ci_integration"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "efb5510d6c82fc01f8984536891b74126f7b0aee"}
{"id": "train_24763", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "repo_scale_reasoning", "tests_are_truth", "reproducibility"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6090f63c611a48be6af45decc29eedb18239d475"}
{"id": "train_24764", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "security_gates", "reproducibility"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d05c6ec4ce06c90d8e49528f1f983d53fd961d1f"}
{"id": "train_24765", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "governance", "documentation", "tests_are_truth"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "766ad2309c65e447a8ed69f087fe3e60ca47e6ec"}
{"id": "train_24766", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "documentation", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "171306bc10e2d9deaa462f277f42b07733c473de"}
{"id": "train_24767", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "documentation", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ef9f54720b38139fb593f707e8d609f875e39234"}
{"id": "train_24768", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "ci_integration", "auditability", "security_gates"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "410c5fa7777ab31d4e6d9b71907707e70aa2c6ef"}
{"id": "train_24769", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "ci_integration", "repo_scale_reasoning", "documentation"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f0619f3888b6e8424c0d08380670c878aa60e48e"}
{"id": "train_24770", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "governance", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "34c25e5e463d9ba480c71c48a561179c8a13df8c"}
{"id": "train_24771", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "auditability", "tooling", "evaluation_metrics"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8bda9a0919de3a90887adccd9eef80fd02997eac"}
{"id": "train_24772", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "documentation", "auditability", "governance"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3c26b05da744f550842b88169082eb993b8cb43e"}
{"id": "train_24773", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "reproducibility", "security_gates", "auditability"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "594e94c7cb3ead7f93d01bc504037e0d6aec6d82"}
{"id": "train_24774", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["governance", "ci_integration", "security_gates", "tests_are_truth"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c4d3b62ab12ac018c3316bba81cf00a0b51446ae"}
{"id": "train_24775", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "evaluation_metrics", "documentation", "governance"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b8bcdedd9d202d4e02525aa497d45fc2f9d80b57"}
{"id": "train_24776", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "documentation", "repo_scale_reasoning", "auditability"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7f6cfec8dbe3be2350a25e6c6c58d4996236b677"}
{"id": "train_24777", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "reproducibility", "governance", "documentation"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d1be3c7876974f15ec968d0a6ab9b30f9e3e7499"}
{"id": "train_24778", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "tests_are_truth", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "18a90d01e0d232e17f92ac6a812aa9c66e7dee17"}
{"id": "train_24779", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "reproducibility", "auditability", "evaluation_metrics"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "91b1c2ab8b8b744f326a8dbe4c35db3214f8b2d1"}
{"id": "train_24780", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "reproducibility", "security_gates", "tests_are_truth"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6043904584e436510444a01e43fcabc1c0915e7c"}
{"id": "train_24781", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "tooling", "repo_scale_reasoning", "evaluation_metrics"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b2240c130e692cc7c383e0e7f98dcbcff4da4744"}
{"id": "train_24782", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "evaluation_metrics", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "40a456c3f154e2e20ca802e880f816bb7b6d8406"}
{"id": "train_24783", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "security_gates", "tests_are_truth"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ced6fc465fe270d8c79c7fdeb4dc79dfe1312284"}
{"id": "train_24784", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "tests_are_truth", "security_gates"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f38e839c7f4e12ea2f2ea3709fc3c87bcb980ba3"}
{"id": "train_24785", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "repo_scale_reasoning", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b15d805506c2b72afbbb2447278759a8e96045cd"}
{"id": "train_24786", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "repo_scale_reasoning", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e13e032488895f8926dd7b5c4dedbb28a4774dc4"}
{"id": "train_24787", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "reproducibility", "evaluation_metrics"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "36ffdeb41d81f9d6af07ea88bdef7bc6788f613b"}
{"id": "train_24788", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "tests_are_truth", "governance"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c1fbadff1a3aa3335455e2433319861960c230ae"}
{"id": "train_24789", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "repo_scale_reasoning", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f7a880eefd411b83573b469b039909bc98892b79"}
{"id": "train_24790", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "evaluation_metrics", "documentation", "repo_scale_reasoning"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "04e1850e2132a168069571c699f41bd9652058b4"}
{"id": "train_24791", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "tooling", "auditability", "evaluation_metrics"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7f61ce3af3cd570fd2d817e84cd50a8df25752c6"}
{"id": "train_24792", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "evaluation_metrics", "security_gates", "governance"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "221d802af38eb1cd86fce9f34f66275c22c32a80"}
{"id": "train_24793", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "repo_scale_reasoning", "security_gates", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ee63f681409f67ea3ac605de89d644a14e865432"}
{"id": "train_24794", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "governance", "ci_integration", "documentation"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "775554de9a4e6d4a3589555a81b6ea3c9f056dc5"}
{"id": "train_24795", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "documentation", "repo_scale_reasoning", "security_gates"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8e5a181d37284f48dcecb74975fa64d4fd3bcb13"}
{"id": "train_24796", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "tooling", "cost_latency_tradeoffs", "governance"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "522c8569dc035f74742e008a1c5ee90c4e45e676"}
{"id": "train_24797", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "auditability", "evaluation_metrics", "documentation"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3eb0a585bdf23ada423d04e42c7155bdbb725592"}
{"id": "train_24798", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "repo_scale_reasoning", "evaluation_metrics", "governance"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "449e8aa7daffdcd92c4a729c9dc89643eab1cd0c"}
{"id": "train_24799", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["governance", "repo_scale_reasoning", "ci_integration", "tests_are_truth"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3a53293a4d826d62e4a560d1ea66c7d57b76498b"}
{"id": "train_24800", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "security_gates", "governance"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8e6c2edfde510dcfd6b47c24afee09a3d49afcc4"}
{"id": "train_24801", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "tests_are_truth", "tooling", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "39a7f8e5afca54ec71608ac5f18fce8df5b950aa"}
{"id": "train_24802", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "security_gates", "evaluation_metrics", "tooling"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "46862f5cfd3eaf4658ffb65cb4f94520fb939e7b"}
{"id": "train_24803", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "auditability", "documentation", "tests_are_truth"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f86599ff0c480d2530546b36911fa44246080dd1"}
{"id": "train_24804", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "tooling", "ci_integration", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2d51949f91ca92607f2f2cc17ae08034ae1c5566"}
{"id": "train_24805", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "cost_latency_tradeoffs", "auditability", "reproducibility"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c3bd8e610cc98a9d7b85a4d01346af84afa3c483"}
{"id": "train_24806", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "security_gates", "repo_scale_reasoning", "documentation"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "97924175fb9c86b1602e1b375602f6ac7e4931c4"}
{"id": "train_24807", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "evaluation_metrics", "governance"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "79f3dfe8aab2df1d277713dccc8d4049d959cd91"}
{"id": "train_24808", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "evaluation_metrics", "ci_integration"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3188df7e7fea22eeb461562eaea7fc77239480d4"}
{"id": "train_24809", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "ci_integration", "documentation", "tooling"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "89f49ca5663d457a8c068ea0cff91202b048b537"}
{"id": "train_24810", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "ci_integration", "governance", "documentation"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c36f91e7b13b5bee8377d46c4737b3e2b44272ce"}
{"id": "train_24811", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "evaluation_metrics", "tooling", "tests_are_truth"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6603fea7bcec7b7f371d5938c33045f1d275970d"}
{"id": "train_24812", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "tooling", "documentation", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fde29419ed05ad893b624be1c9c01d4298fcaa69"}
{"id": "train_24813", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["governance", "auditability", "security_gates", "documentation"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6ef1a6ae8dbe1c6ea7e3556b8d1a1153684ef092"}
{"id": "train_24814", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "ci_integration", "documentation"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "90ef92d0f50336c1d1624b0fb09dddf73cef6452"}
{"id": "train_24815", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["governance", "security_gates", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bd56a260359f4b8f64bc68af3668880d171ee71d"}
{"id": "train_24816", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["governance", "security_gates", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "65590d8764e157cb6906e7350d831bd0ef2ea541"}
{"id": "train_24817", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "tooling", "security_gates", "ci_integration"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "33b89348338c3fd76334ecb77f5f7680b3a40239"}
{"id": "train_24818", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "tooling", "documentation", "security_gates"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "34c25e5e463d9ba480c71c48a561179c8a13df8c"}
{"id": "train_24819", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "reproducibility", "documentation", "governance"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7ad73270d456fe90d4a78c60974bc1a2f9def578"}
{"id": "train_24820", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "evaluation_metrics", "repo_scale_reasoning", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "47f449d69efd9bc6035542d4e8409ae976cbc8a5"}
{"id": "train_24821", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "ci_integration", "security_gates", "tests_are_truth"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a8a3b6a08f942e90ce72a10a00b5232624efea48"}
{"id": "train_24822", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "governance", "evaluation_metrics", "ci_integration"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "82178fed697e5836e7d8db21fcbb0ff04540679b"}
{"id": "train_24823", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "evaluation_metrics", "ci_integration", "repo_scale_reasoning"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4e869699479845e0b9483e8e505dd6d788b670b9"}
{"id": "train_24824", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "evaluation_metrics", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fd095c7101a79631125be38d3d44a3ae251176d1"}
{"id": "train_24825", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "auditability", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f04116d361708962a39ae4885c62243b98e8b9e2"}
{"id": "train_24826", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "security_gates", "evaluation_metrics", "documentation"], "moe_experts": ["security_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cfa3d25a8ebf91f80f1eb8ce936ba36d378bb945"}
{"id": "train_24827", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "reproducibility", "ci_integration"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "264325d08617ec80c17bd19e4e19ce9d9dc1ab9e"}
{"id": "train_24828", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "evaluation_metrics", "repo_scale_reasoning", "tooling"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3114aace7c00041594bf28407d9f3b107a44f87c"}
{"id": "train_24829", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "reproducibility", "security_gates"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b601357f95d21ad5ed49bd59d71627125e2f379f"}
{"id": "train_24830", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "security_gates", "governance", "tooling"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2987fcfd08395894c82948be8f8310e60a87cbb3"}
{"id": "train_24831", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "reproducibility", "auditability"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2221df414100b0556f01576d6ae1b9fee4be5238"}
{"id": "train_24832", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "tooling", "evaluation_metrics", "governance"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "de1416fb3aaecb5f3e631a30c513e2f45bfbafd1"}
{"id": "train_24833", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "ci_integration", "tooling", "tests_are_truth"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c0238ec9c795d82925ef8bcfca5442d5d9ac437a"}
{"id": "train_24834", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "tooling", "reproducibility", "security_gates"], "moe_experts": ["security_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3b8c6873ccefc8d36643a1f441c64cdee3556153"}
{"id": "train_24835", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "ci_integration", "cost_latency_tradeoffs", "tests_are_truth"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8e87bc0fd39b1875726db13d278c6e891cd39265"}
{"id": "train_24836", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "security_gates", "documentation", "tooling"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "12dacf9c10c8907793eda196e2e8440d3bffce6e"}
{"id": "train_24837", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "auditability", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4002e72ff986db440244ae1eee40d7b2c0774ccf"}
{"id": "train_24838", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "security_gates", "governance", "auditability"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9268b64a3fe1c414af77d1d4d27118c16d29254f"}
{"id": "train_24839", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "auditability", "documentation"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3c26b05da744f550842b88169082eb993b8cb43e"}
{"id": "train_24840", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "security_gates", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["performance_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "68a7d4df2ea630bf4d67eeb86f04890751da95b6"}
{"id": "train_24841", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "ci_integration", "auditability", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "701c25faed46b84d25846bc0cc03719736d1ddf8"}
{"id": "train_24842", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "governance", "documentation", "tests_are_truth"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e9e27de1da7d8c0fee5b161f247d13da873f7045"}
{"id": "train_24843", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "ci_integration", "tooling", "reproducibility"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fa746e7bd977070f248b35b2ddbe6463303b97e0"}
{"id": "train_24844", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["auditability", "documentation", "evaluation_metrics", "tests_are_truth"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0d8fe929e9d185977c6671a8c8687f5313e0a369"}
{"id": "train_24845", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "tests_are_truth", "evaluation_metrics", "governance"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fc62b09014f22a12a3a2bb5343f72c4f468482b0"}
{"id": "train_24846", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "repo_scale_reasoning", "cost_latency_tradeoffs", "documentation"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "36f4f864dfaa606955bc52885c7784ca154bfdb0"}
{"id": "train_24847", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["governance", "ci_integration", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6b4e8c2688f75a57411ff6590f1075f5ff9f46bb"}
{"id": "train_24848", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "governance", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c4d413f4e61b5d8b8de792bbbfbf1eda115996ad"}
{"id": "train_24849", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "tests_are_truth", "evaluation_metrics", "documentation"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fe5375551eb42c69523d48b79406df3adef15e85"}
{"id": "train_24850", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "auditability", "tooling"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5868012c3106737bc36b4c9e99fc4030643198a5"}
{"id": "train_24851", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "tooling", "governance", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d58af9b5d92e5134630b1f9af394a3dd164cb0b6"}
{"id": "train_24852", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "tooling", "security_gates", "tests_are_truth"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "80fdb98b1679f56dc9da7e74eb7d992b40ef691a"}
{"id": "train_24853", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "tooling", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ea48fd2e3d24b4aa255ee59d38f6c759a8c121d0"}
{"id": "train_24854", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["governance", "documentation", "repo_scale_reasoning", "auditability"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "407b9fb72debba701439306af977e0718726e2b2"}
{"id": "train_24855", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["governance", "auditability", "ci_integration", "tooling"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "02bbd351d685048a1e645d3f8dcc28e10cb8e18c"}
{"id": "train_24856", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "governance", "tests_are_truth"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "350c3adf77ac4d51c5008cd66a4f8792b2f19f43"}
{"id": "train_24857", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["governance", "tooling", "ci_integration", "security_gates"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "310739694b1ce7ac542a0dec38f77431f8cafb1f"}
{"id": "train_24858", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "cost_latency_tradeoffs", "evaluation_metrics", "ci_integration"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "275aa9ed4857cda79904dd0ad02cb02d725bb106"}
{"id": "train_24859", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "governance", "security_gates"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8b5239705efeea75686c312f7359dffc757f2746"}
{"id": "train_24860", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "ci_integration", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "de770cc413c6bb909a87a932f221d06c129e7091"}
{"id": "train_24861", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "auditability", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8244be2c74c47c9481618f17a570e73c31ba2892"}
{"id": "train_24862", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "tooling", "cost_latency_tradeoffs", "governance"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "eafe37a40e7dcdc45ed76e98be4d8b74401bd9b9"}
{"id": "train_24863", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "cost_latency_tradeoffs", "documentation", "auditability"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b9d391edaf886bdd3167de0a15e5f6f8704f94be"}
{"id": "train_24864", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "auditability", "reproducibility", "tests_are_truth"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6837cc9a349497ee09c6d81868c06eb27517451d"}
{"id": "train_24865", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "reproducibility", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "97ed7f19dd23e39a3e713d62421598b60d8bf57d"}
{"id": "train_24866", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "tooling", "evaluation_metrics", "ci_integration"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2734c2e17874e46a8c069c00ce1c676b20be5a9b"}
{"id": "train_24867", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "evaluation_metrics", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "987f5e1f9d72becefdf43c0b2a1c85a29fe10025"}
{"id": "train_24868", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "evaluation_metrics", "reproducibility", "tooling"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6e8994b925c191aa94fc9c158a576d902d9c6e73"}
{"id": "train_24869", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["governance", "evaluation_metrics", "reproducibility", "tooling"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "068631152cec9ab516d5da9d29c5bef76aa0f9bd"}
{"id": "train_24870", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "tooling", "governance"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "577230705a404d0a3d095e6941524c463c377f1d"}
{"id": "train_24871", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "ci_integration", "tests_are_truth", "tooling"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "56971a0893d575866df557ef452ba2c6dd9e29ac"}
{"id": "train_24872", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["auditability", "governance", "evaluation_metrics", "ci_integration"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cc782cf9449e9def006de9ddf99ec041a1e9f46f"}
{"id": "train_24873", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "tooling", "repo_scale_reasoning", "security_gates"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ac73df952c088e4ee7b542f6b70bc619dcca3076"}
{"id": "train_24874", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: C#\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "auditability", "security_gates", "governance"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1bf5b72b8ece6dbb38c13cf171e23c6a212ae4fe"}
{"id": "train_24875", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["governance", "evaluation_metrics", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "15d222a9a167b1b4d13d5ecbe35bf269eb906582"}
{"id": "train_24876", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "security_gates", "governance", "reproducibility"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6349b401c4eed74fa0091947277346483ce80992"}
{"id": "train_24877", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "tests_are_truth", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "753de58755ce940f67ba7fb0c7c397da833d6874"}
{"id": "train_24878", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["auditability", "tooling", "reproducibility", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "eab4942071498efc0c212c8f7a8c329ed916fda9"}
{"id": "train_24879", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "governance", "reproducibility", "auditability"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c3ff5b71975621607ae4e19ecfd5691ed890feed"}
{"id": "train_24880", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "tests_are_truth", "evaluation_metrics", "documentation"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9ee57b681627ae496424a5edda145f5e39030a09"}
{"id": "train_24881", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "security_gates", "tooling", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b549417db1ae5bd501c8f07456422f7a2841d5de"}
{"id": "train_24882", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "auditability", "evaluation_metrics"], "moe_experts": ["coding_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8b7e2f2d2e7330408a9021e5f70702fc76cbb240"}
{"id": "train_24883", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "ci_integration", "governance", "auditability"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3af7ecf7c569fe65c824205cfa14158744255790"}
{"id": "train_24884", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "tooling", "auditability"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f2dcc32d3defbe8d98f118d74360172068ba61b1"}
{"id": "train_24885", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "governance", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2be7a5d0c12134f1d88a2e6355895cae45972ffb"}
{"id": "train_24886", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "reproducibility", "auditability", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c14222e2d95ab5a53e0920ca5d8372680e4dfaf3"}
{"id": "train_24887", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "reproducibility", "tooling", "tests_are_truth"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d87e46793fe013655c00001299bab1b929455a64"}
{"id": "train_24888", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Python\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["auditability", "tests_are_truth", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "31bb34a20aae8a557c9cb451f4d20abe55c3ec6d"}
{"id": "train_24889", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "ci_integration", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0f159440d4cb7886c81408b24d9b208a660afbd7"}
{"id": "train_24890", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "repo_scale_reasoning", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "905ce2261772d7f2029fea94179f642912b9d8c6"}
{"id": "train_24891", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "ci_integration", "auditability", "security_gates"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "69302c240fd0ccca6d084b4a907877c45c8a4a1d"}
{"id": "train_24892", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["governance", "security_gates", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e656538733914f1326033da33fbbc92ae9247caf"}
{"id": "train_24893", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "auditability", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bc628b513931eb68b8d4a51b66c6ddf592b54574"}
{"id": "train_24894", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "reproducibility", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "244a38f840d6656f7a9abc0762fcf2a5bb90209e"}
{"id": "train_24895", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["auditability", "evaluation_metrics", "documentation", "reproducibility"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "856b6df97380920c0117bd99e5d54d30bc9a7988"}
{"id": "train_24896", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "documentation", "tooling", "evaluation_metrics"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "01c2a54f043f06b40f89bd010dc8cb9ffb3abd61"}
{"id": "train_24897", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "ci_integration", "security_gates", "tests_are_truth"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "321f9289f944436fcd4a53cbdb230c5ff5c8b258"}
{"id": "train_24898", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "security_gates", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "3c57bac22934e537a6a60cc93b0c9f5855f41cc9"}
{"id": "train_24899", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "reproducibility", "ci_integration", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5d14e06e7c3067f428f82a5f10e854d30f1d7ff3"}
{"id": "train_24900", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "tooling", "reproducibility", "governance"], "moe_experts": ["evaluation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d0789c2d25fa9089e95e42b3e62a37e83ca3801e"}
{"id": "train_24901", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "repo_scale_reasoning", "reproducibility", "security_gates"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a33ec930f676ba31b02da81b36311551d69fd1fa"}
{"id": "train_24902", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "ci_integration", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a1ff4289366ab49b380e5f80de71ef692911cb15"}
{"id": "train_24903", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["governance", "repo_scale_reasoning", "ci_integration", "reproducibility"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4d6f669eb9ae148417aa7f452c7d0c3d0878ed21"}
{"id": "train_24904", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "evaluation_metrics", "security_gates", "auditability"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2059964c637caf2d68cb6c1051efcfe3cd2a4078"}
{"id": "train_24905", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "governance", "documentation", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f2d355d876419d0742b2f1413956d51194db6e45"}
{"id": "train_24906", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["governance", "ci_integration", "auditability", "tests_are_truth"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "110d737e656f4dae71f2d7ee9d55001d75d154ab"}
{"id": "train_24907", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["auditability", "documentation", "tests_are_truth", "governance"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "316f6530aecdb63b553a6095e68694f6c9f04569"}
{"id": "train_24908", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "reproducibility", "governance", "auditability"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5c77e79343dbbe4337c29aa40f47bc36b91152b7"}
{"id": "train_24909", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["governance", "tests_are_truth", "tooling", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a36d5375b64a9686abd0dc5dd8868e30cc029005"}
{"id": "train_24910", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "ci_integration", "tooling", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cca2cd25801f553288fc75720688d046f9916be5"}
{"id": "train_24911", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "tests_are_truth", "security_gates", "ci_integration"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cb65172e824691ebaf82f7696b5638687659cda1"}
{"id": "train_24912", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["auditability", "repo_scale_reasoning", "evaluation_metrics", "ci_integration"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "ae3aecb9aab3a4be501ca0198cc1824f2459bc11"}
{"id": "train_24913", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "ci_integration", "tooling", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cb2185cfe0fa067a1bdb23bf309fbc95fc8007d1"}
{"id": "train_24914", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "documentation", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["governance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e60a641ca130837a8025326e7acafa36bfe2bb4b"}
{"id": "train_24915", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "documentation", "governance"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6d9bd54e775dde0812f6720662b6a1e655c5d35d"}
{"id": "train_24916", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["governance", "cost_latency_tradeoffs", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "61337214327c3d2caf5f228b0a487dc508f4c76b"}
{"id": "train_24917", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "governance", "documentation"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a5e1ddd4f579f827d243fce64a6f17064229b85f"}
{"id": "train_24918", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "auditability", "reproducibility", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "51e81ae8461c1018cb228093e9b81dd14889c720"}
{"id": "train_24919", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "tooling", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5998b0a376ee36093ed8552bddd786e47302e07f"}
{"id": "train_24920", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["documentation", "ci_integration", "governance", "tests_are_truth"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e7351023c1276d78271ed2b18045761a6fa7e510"}
{"id": "train_24921", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "auditability", "documentation", "tooling"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "77461fd72bb73d847f12599e19be50f4a6da4444"}
{"id": "train_24922", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "ci_integration", "tooling", "auditability"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5069932daa1bc90f0d208b27af2e95e039cea86c"}
{"id": "train_24923", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "reproducibility", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "afc9946ba38105fbf9d571ab24ef8831bd28ac05"}
{"id": "train_24924", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "tests_are_truth", "reproducibility"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "89824fdb4d3b54ce83cdfa9535098a07dae0945c"}
{"id": "train_24925", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "tests_are_truth", "reproducibility", "documentation"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e8cd808037595dccdd39861869af3869edd0ed16"}
{"id": "train_24926", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "auditability", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "dee4bc85af11249ce86c90a3d6112ab529f65bf0"}
{"id": "train_24927", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "reproducibility", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["data_curation_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e27831f9824e6bb018889a111385fe1ce0706aee"}
{"id": "train_24928", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "repo_scale_reasoning", "tooling"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a40b044a81be35a72b78015613813477b20ea7cb"}
{"id": "train_24929", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: SQL\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "tests_are_truth", "evaluation_metrics", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "458dd9a67482154c63dc272ed7d90cd1832b1e2d"}
{"id": "train_24930", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Self-improving agents and feedback loops\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "evaluation_metrics", "governance", "ci_integration"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "499fe49ef0457fb05f0aa11246c9ec9c4b066a7c"}
{"id": "train_24931", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "ci_integration", "tooling", "repo_scale_reasoning"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "72171cfd5d89028896407b50087db71fa2493874"}
{"id": "train_24932", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "repo_scale_reasoning", "cost_latency_tradeoffs", "evaluation_metrics"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cd50f34c7aea72db1dddda25dabbc5982a07ad5f"}
{"id": "train_24933", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "auditability", "tooling", "security_gates"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d6053af3a6aad0f26a851e245b7a1d7e957f16f6"}
{"id": "train_24934", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Bash\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "repo_scale_reasoning", "reproducibility", "security_gates"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cc0218bebc76b581bc0f418fb0d0f8c304a69f6d"}
{"id": "train_24935", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Go\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "ci_integration", "tooling", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2825d731493a60d992efd044439558a44c9b5b66"}
{"id": "train_24936", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["auditability", "ci_integration", "governance", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b4d60f6da857d2faf4ce01455d5956d5a08e2cee"}
{"id": "train_24937", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Java\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["governance", "tests_are_truth", "ci_integration", "documentation"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "76c486973ad2a80f8cff8651ae10be503b24250f"}
{"id": "train_24938", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "auditability", "ci_integration", "governance"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b637b630830d96793fc8d3bebbdea2b3e2333b89"}
{"id": "train_24939", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["auditability", "tooling", "evaluation_metrics", "repo_scale_reasoning"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f0845f2a28175f2e271dad5b90d2c5bdf8d0f122"}
{"id": "train_24940", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: TypeScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "tests_are_truth", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d53ced52f1a7ec83bbf33716ab5bb8241f0c9a76"}
{"id": "train_24941", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "reproducibility", "tooling", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "e7e08bdc195a503699eba9e3f0b8458b77855531"}
{"id": "train_24942", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: JavaScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "documentation", "tooling"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c7c95a5efbcca3a5e4be2a44c2d78364663a467f"}
{"id": "train_24943", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "reproducibility", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "f93eb61c920f7db195d6bfb3c0785bf09e15e9fc"}
{"id": "train_24944", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "repo_scale_reasoning", "auditability", "tests_are_truth"], "moe_experts": ["evaluation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0289e9f97c1e9e5f98c6b90c76401acfc51d9471"}
{"id": "train_24945", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "documentation", "tests_are_truth", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "539e7df81984e833ac62d7c25e3c5032873ef730"}
{"id": "train_24946", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "evaluation_metrics", "tooling"], "moe_experts": ["governance_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a2537134fcdd164ac33c038ea896758ae1ddec66"}
{"id": "train_24947", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "JavaScript", "developer_needs": ["auditability", "security_gates", "documentation", "reproducibility"], "moe_experts": ["security_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0930f012a46ea5df3989d73625104fbc114bf1c4"}
{"id": "train_24948", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "evaluation_metrics", "cost_latency_tradeoffs", "reproducibility"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "17a00d77213653cf9648f10a4ed8d5d52784efc4"}
{"id": "train_24949", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["auditability", "governance", "tests_are_truth", "documentation"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7f6e4dd51d4268f5c6c49e5a6669439c10574999"}
{"id": "train_24950", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "auditability", "evaluation_metrics", "tests_are_truth"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "8a2c3050a22493112f870ca05284ae1467ae9e9e"}
{"id": "train_24951", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: TypeScript\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "reproducibility", "tests_are_truth", "security_gates"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "97f979f9517da9b062e1eacf2fb865701573ca7f"}
{"id": "train_24952", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Bash\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "reproducibility", "security_gates", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c2173b0612e4bd102a60fbcecefbc511ea0b868c"}
{"id": "train_24953", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Java\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "governance", "tests_are_truth"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "665847072c463a70c415b8e6317cbf42b8fbb0e0"}
{"id": "train_24954", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "security_gates", "repo_scale_reasoning"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1155cf217b74f8a0a64b9561d763879209265c2f"}
{"id": "train_24955", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "security_gates", "tests_are_truth", "cost_latency_tradeoffs"], "moe_experts": ["coding_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fd65b606668c09e62e8a762730db426a6db22741"}
{"id": "train_24956", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "security_gates", "governance", "tooling"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "98c5cedec94520c4a572f94e33ac9d04b12f0615"}
{"id": "train_24957", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Latency, cost, and reliability optimization\nDifficulty: advanced\nTarget language: Rust\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "auditability", "documentation", "cost_latency_tradeoffs"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "622eb129d0b2ca1c67f1247696ae5345a876bfec"}
{"id": "train_24958", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["governance", "tests_are_truth", "documentation", "security_gates"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "df146e49562dbcd135193a7957623d9ffabb804c"}
{"id": "train_24959", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "ci_integration", "tests_are_truth", "evaluation_metrics"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "97ed7f19dd23e39a3e713d62421598b60d8bf57d"}
{"id": "train_24960", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: SQL\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "auditability", "documentation"], "moe_experts": ["coding_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "28c7933ad3f8fa1a5c88470ca5d2fef7d60b51d0"}
{"id": "train_24961", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Go\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "auditability", "cost_latency_tradeoffs"], "moe_experts": ["evaluation_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fb91e9cc17b75d690bdc40473808ef118719e355"}
{"id": "train_24962", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: Java\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "tooling", "security_gates", "reproducibility"], "moe_experts": ["security_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c8f5401e9d327aac096856adbe3ab1918995fa71"}
{"id": "train_24963", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "auditability", "reproducibility", "tooling"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "92994400e6e5d57b097b963e60cbb9a811ecd520"}
{"id": "train_24964", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: C#\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["auditability", "evaluation_metrics", "reproducibility", "tooling"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "db71a26db7f47ffd4e74a1893859906cf1d6766e"}
{"id": "train_24965", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: TypeScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nReview: correctness, security, performance, governance\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "tooling", "cost_latency_tradeoffs", "auditability"], "moe_experts": ["governance_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "b85e533e1c5326612e7b9127621a4956977c4592"}
{"id": "train_24966", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Rust\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "ci_integration", "auditability", "security_gates"], "moe_experts": ["data_curation_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9d3510e4636ab343615ba1b37843241c5cf077c0"}
{"id": "train_24967", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: TypeScript\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "ci_integration", "reproducibility", "governance"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "08b9662a51a4624e60a4c2f7fb2c384b194620cc"}
{"id": "train_24968", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "governance", "ci_integration", "documentation"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "01370dfcbb2f5267610aad90a9dd8f22a8dedffc"}
{"id": "train_24969", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "auditability", "governance", "security_gates"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "42f83726e41c31609abade0404143e06499f04a5"}
{"id": "train_24970", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: C#\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["tooling", "tests_are_truth", "documentation", "governance"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d3cc84c0a7141bdcc758b2c29322e923a93e56f8"}
{"id": "train_24971", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Python\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "security_gates", "documentation", "ci_integration"], "moe_experts": ["agentic_systems_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "46d7ac99a983bc079a0b3154059d27802676a2d4"}
{"id": "train_24972", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Bash\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "cost_latency_tradeoffs", "repo_scale_reasoning", "governance"], "moe_experts": ["data_curation_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "a22a9a7eb4bcdfe8f5de5b8f9e435cf151975ad0"}
{"id": "train_24973", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Go\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["tooling", "reproducibility", "repo_scale_reasoning", "tests_are_truth"], "moe_experts": ["performance_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "779abec667855149c4e5668621f5ea5629e43138"}
{"id": "train_24974", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: TypeScript\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "governance", "repo_scale_reasoning", "reproducibility"], "moe_experts": ["evaluation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d5a42477005db143bc9b083c1319ed53ebef360e"}
{"id": "train_24975", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "governance", "ci_integration", "tooling"], "moe_experts": ["data_curation_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "483800a9d253a3713e5029f3abe793afc27a21f9"}
{"id": "train_24976", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nTarget language: Go\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "governance", "cost_latency_tradeoffs", "tooling"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "bc91147bfcc886e6845c7282b45e475a7862a10d"}
{"id": "train_24977", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Java\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["documentation", "ci_integration", "tooling", "repo_scale_reasoning"], "moe_experts": ["data_curation_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "20b84484fb4f8de45443f72004017c6bda7e1c26"}
{"id": "train_24978", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "patch_diff", "difficulty": "intermediate", "instruction": "Task: patch_diff\nTopic: Latency, cost, and reliability optimization\nDifficulty: intermediate\nTarget language: Python\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["governance", "auditability", "repo_scale_reasoning", "ci_integration"], "moe_experts": ["coding_expert", "performance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "62c6872f057c38a654032d9894a45c83bdeae1e2"}
{"id": "train_24979", "created": "2026-01-01", "topic": "Latency, cost, and reliability optimization", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Latency, cost, and reliability optimization\nDifficulty: expert\nTarget language: JavaScript\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "documentation", "evaluation_metrics", "cost_latency_tradeoffs"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cff9c66a6c0e6d88f4b9259225202471708295ad"}
{"id": "train_24980", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Rust\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "auditability", "evaluation_metrics", "tests_are_truth"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "c4ffa8b3d0a41614b09e3ee915f76fe256a7d03f"}
{"id": "train_24981", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Rust\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "documentation", "reproducibility"], "moe_experts": ["performance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "07154649c872c9b6ad13a62b9107e23cd06cc955"}
{"id": "train_24982", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Self-improving agents and feedback loops\nDifficulty: intermediate\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "C#", "developer_needs": ["documentation", "cost_latency_tradeoffs", "governance", "evaluation_metrics"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5b81edeeece53dbaa70c82537574207c17bd6b6f"}
{"id": "train_24983", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Java\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": [], "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "tooling", "governance"], "moe_experts": ["security_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "6b39545ffcf3c1d09fb8e84176babea0d096a3ea"}
{"id": "train_24984", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Bash\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nLoop: Plan \u2192 Edit \u2192 Test \u2192 Reflect \u2192 Human gate\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "auditability", "tooling", "security_gates"], "moe_experts": ["security_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "0c1196f7925993479fe92b7cb8e861b71fe4bddd"}
{"id": "train_24985", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "intermediate", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nFailure:\n- Initial patch broke edge case\nReflection:\n- Missing zero-input guard\nCorrection:\n- Add explicit validation + test\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "Python", "developer_needs": ["tooling", "documentation", "tests_are_truth", "governance"], "moe_experts": ["performance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "9bf6f3cedbffd8681ab7105b2201fce0fc102d1a"}
{"id": "train_24986", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "tests_are_truth", "documentation", "governance"], "moe_experts": ["agentic_systems_expert", "security_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "2dbd35e026f22cbd0486ef8881c2069a1f338b9e"}
{"id": "train_24987", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nTarget language: C#\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "repo_scale_reasoning", "security_gates", "tooling"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cb1ded83e66eacd2de38957cd00f2c1fec0b2d2c"}
{"id": "train_24988", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: SQL\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPatch (diff-style):\n```diff\n- if x == 0:\n-     return 1/x\n+ if x == 0:\n+     raise ValueError('division by zero')\n```\nAcceptance:\n- Tests pass\n- No new regressions\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "documentation", "tests_are_truth", "auditability"], "moe_experts": ["agentic_systems_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "177666c42d2e51b87132a09de8451d487fa27ca7"}
{"id": "train_24989", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Python\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "governance", "auditability"], "moe_experts": ["performance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "fea2debfb4d82f9077bbe128ed0dc65712f2460f"}
{"id": "train_24990", "created": "2026-01-01", "topic": "Self-improving agents and feedback loops", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Self-improving agents and feedback loops\nDifficulty: advanced\nTarget language: JavaScript\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": ["Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions"], "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "auditability", "documentation"], "moe_experts": ["governance_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1f8d1748afa01f667033fef030bd2e05693a1fdb"}
{"id": "train_24991", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: C#\nContext: Regulated environment requiring audit trails.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "tooling", "cost_latency_tradeoffs"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "7a8cea2a0ed03acb9d0b7c69bdac9c12b39a9865"}
{"id": "train_24992", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Bash\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "tests_are_truth", "auditability", "reproducibility"], "moe_experts": ["coding_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "d27aebe97df48fe4ef3a6f0c777e405b064f9f99"}
{"id": "train_24993", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "security_gates", "ci_integration", "governance"], "moe_experts": ["agentic_systems_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "4b532327361221961dbb90b46ba2a679067b8eff"}
{"id": "train_24994", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: SQL\nContext: Offline/local deployment with limited compute.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nPipeline:\nIngest \u2192 Normalize \u2192 Filter \u2192 Dedupe \u2192 Quality \u2192 Mix \u2192 Audit\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "tooling", "documentation", "repo_scale_reasoning"], "moe_experts": ["agentic_systems_expert", "data_curation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "19593c1195bf41473cdcf60bca41e20869a6784b"}
{"id": "train_24995", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Rust\nContext: Research team validating claims against real repos.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nEval:\npass@k, time-to-green, regressions, diff size\n", "reasoning_steps": [], "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "auditability", "tooling", "security_gates"], "moe_experts": ["agentic_systems_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "621e216755762b386764589003aa4d175edfbb51"}
{"id": "train_24996", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Python\nContext: Large monorepo with flaky tests and strict CI.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nScaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=4):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"ok\": ok})\n        if ok:\n            return patch, history\n        p = p + \" | refine\"\n    return patch, history\n\n```\n", "reasoning_steps": [], "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "tooling", "evaluation_metrics"], "moe_experts": ["governance_expert", "coding_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "1a9667fe81df646d88cb52f23b96f783fab464bc"}
{"id": "train_24997", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "reproducibility", "cost_latency_tradeoffs", "ci_integration"], "moe_experts": ["data_curation_expert", "governance_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "001c6f647411889a75d75ea1c41d2019b08ffae4"}
{"id": "train_24998", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: SQL\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nDesign with risks, metrics, acceptance criteria\n", "reasoning_steps": [], "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "documentation", "governance"], "moe_experts": ["governance_expert", "evaluation_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "cc673dc256410ae5a58241d001a3639fb5181b9b"}
{"id": "train_24999", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nTarget language: Go\nContext: High-traffic service with latency SLOs.\nProduce expert-level, production-ready artifacts.", "input": "", "output": "Facts:\n- Modern AI coding prioritizes correctness, evaluation, and governance.\n- Agentic loops with test gates outperform single-pass generation.\n\nCompare: capability, cost, latency, reliability\n", "reasoning_steps": [], "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "security_gates", "evaluation_metrics"], "moe_experts": ["coding_expert", "agentic_systems_expert"], "governance": {"audit_required": true, "tests_required": true}}, "hash": "5db16b23daf471e1921f948e241d83959cf1391c"}